Good morning. Welcome to AIG Investor Day 2025. Today's remarks may include forward-looking statements, which are subject to risks and uncertainties. These statements are not guarantees of future performance or events, and are based on management's current expectations and on assumptions currently believed to be reasonable.
AIG's filings with the SEC provide details on important factors that could cause actual results or events to differ materially. Any forward-looking statement made during this presentation speaks only as of the date on which it is made. Except as required by applicable securities laws, AIG is under no obligation to update any forward-looking statements if circumstances or management's estimates or opinions should change. Today's remarks may also refer to non-GAAP financial measures.
The reconciliation of such measures to the most comparable GAAP figures is included in our earnings release, financial supplement, and earnings presentation, all of which are available on our website at aig.com. Finally, today's remarks related to General Insurance results, including key metrics such as net premiums written, net premiums earned, losses and loss adjustment expense incurred, underwriting income, margin, and underwriting leverage, are presented on a comparable basis, which reflects year-over-year comparison adjusted for the sale of Crop Risk Services and the sale of Validus Re as applicable.
We believe this presentation provides the most useful view of our results and the go-forward business in light of the substantial changes to the portfolio since 2023. Please refer to the non-GAAP reconciliation section of the presentation for reconciliations of the metrics reported on a comparable basis. Ladies and gentlemen, AIG Investor Day 2025 is about to begin. Please prepare to join us. Ladies and gentlemen, please welcome Chairman and CEO Peter Zaffino.
Finally, we get to present. Welcome, everybody. It's great to have you here with us today. We're really excited to take you through our story. I do want to just do quickly the agenda, what we're going to take you through today. I'm going to open up and give you some perspective on what's happened at AIG with a little bit more color.
Keith Walsh, our CFO, is going to get up and give us some of the financial information to support that and give you a little bit of view in terms of the future. We'll stop there and do a little Q&A. We'll take a break. We're going to bring up our top business leaders for Q&A and talk more deeply about the business across the world. We'll then pause and start to shift to AI.
We're going to have Claude Wade, who's our Chief Digital Officer, Head of our Business Operations and Claims, take us through this. He's been a pivotal part of our journey. We have an unbelievable panel discussion with Dario Amodei, Alex Karp, Anthropic, Palantir, narrated, and worked through with Sara Eisen. We're going to be really excited to take you through that.
They are our partners, and we'll show you that. They are incredibly important to us, but also a validator in the work that we've been doing there. All of the presenters will come up, and we're going to do Q&A. Let me talk about where this begins. As many of you know, I joined AIG in September of 2017 after spending nearly 17 years at Marsh & McLennan Companies.
When I was at Marsh & McLennan, I worked on AIG my entire time there. I thought I knew the company well before coming to AIG. What I would find out as I was here for my first 100 days is I really did not know the company at all. As I started to do business reviews, deep operational reviews, I learned a lot. The problems that existed were not hard to identify, just because there were plenty.
The bigger ones were there was no underwriting culture. The underwriting discipline was not as strong as we would have liked. We had massive aggregation and exposure issues that needed to be dealt with. There was no end-to-end operational capabilities, and we had no insight really in our data. There was no consistency in our data.
We had the highest expense ratio and the bottom decile of all of our peers, with no real path to kind of get to a place where we would have a better expense ratio. By the way, we had not visited south of a 100 combined ratio in a very, very long time. It reminded me of a story where a young boy went to President Kennedy and asked him a question and said, "How did you become a war hero?" President Kennedy told the boy, "I had no choice. They sunk our boat." AIG had a choice, and AIG's choice was to swim. We needed to restructure the entire company. We needed to elevate the talent level because the path we were about to embark on was unprecedented.
To try to put a company back in place that was so vital and important to the global insurance industry as a top performer. I took our top 100 leaders in early 2018 away to talk about the path in which we had to journey and what was going to be in front of us. It was really important, and I told them it was going to be one of the hardest 24 months they've ever had in their careers.
They're going to work incredibly hard. We're going to make excellent decisions, but we're not going to have any financial results to show for it for at least 24 months. When we bend the curve, and we will bend the curve, we're going to be unstoppable and we'll never look back. I remember telling them, "There's three people you don't want to negotiate with or trade with.
One, somebody who's smarter than you. Two, is somebody who's a better negotiator. Three, us, somebody with nothing left to lose. We began, and we talked very detailed about what was required for this. Over the next three hours, my colleagues and I want to take you through our journey of nothing left to lose. I do not use these words lightly. We have done an unprecedented turnaround, and we feel we have unparalleled opportunity in front of us, and we want to outline that for you today. I cannot take the virtual flip into the back page for guidance, so here it is. By the way, I will do it at the end as well. We are very committed to the financial guidance that I am providing now.
Over the next three years, on a compound annual growth basis, so it may not be one year, but it will be all three, we are going to have 20% plus EPS growth. The core operating ROE will be 10% in 2025, but believe over the next three years, the range is 10-13%. We will continue our path on expense reduction through operational excellence, and other variables we will take you through today. It will be below 30%.
Our board intends to support dividend increases over the next two years of 10% for 2025 and 2026. I think that is a lot of guidance. It is ambitious, but we will achieve it. Six parts I want to take you through today, and it will go by fast. How we revived our brand of underwriting and operational excellence, our reinsurance strategy and why I think you really need to know what you're doing in order to maximize value, an overview of our global businesses, how we've exhibited operational excellence as a core competency. I will give you my perspective as a business leader and the CEO of Driving Gen AI and how we're going to unlock unparalleled opportunities in front of us.
How we're going to relentlessly continue top quartile performance and give you a little bit more perspective, not only on the path, where we are, where we're going. Let me talk about our brand and our underwriting and operational excellence. This is a snapshot, actually quite a bit of information on this page of what our global businesses look like. $24 billion in net premium written.
International commercial and North America commercial, you can see very balanced, 35% of that overall portfolio. Global personal makes up 30%. Really strong geographic diversification. We're in the big countries where insurance is prominent. United States and North America, we are $8.5 billion. You can see in international, U.K. with our global specialties, Talbot in our domestic business, over $4 billion. Japan, we're the largest non-domestic insurance company in the country.
We have strong presence in Asia-Pacific and in Europe. We have the capabilities in over 200 countries for our global clients and multinational, which is significant, and it gives us an advantage to service across the globe, and we'll talk about that. What's on the bottom is the amount of change that's happened in the company. This is something that you don't typically see in companies. This is a significant amount.
Since 2017, 100% of my leadership team is new. 80% of the top 100 leaders are new. Our underwriters, which have done a phenomenal and exceptional job, 67% of the 3,100 underwriters we have are new. The backbone of our company, led by Julie Chalmers, who's here today, 4,200 claims professionals, 45% are new. What would I draw from that? We're a new company.
I mean, we have the same name, but there are so many differences that we want to outline today, and you can see the dramatic change. This is how we break it out by segment. 48% is commercial property and casualty. You see we have sizable businesses that exist within P&C. We're the world's largest commercial specialty underwriters with our global specialty business out of London, led by Gordon Brown. Lexington, led by Lou Levinson. We have our syndicate, Talbot.
We'll talk about global personal. Within that, we have a tremendous global A&H business, homeowners, auto. We've talked a lot about high net worth. We'll go into that a little bit as well. It's a really good balance. Let's look at the journey of what we had to do. 121% combined ratio. These are restated for apples to apples as though parent expenses were fully allocated like they are today.
There are a few things I want to draw your attention to. 121%, we all know, is bad. I mean, it was 83% on the loss ratio, 38% on the expense ratio. That's a hard combined ratio to sort of drive outcomes from. If you look all the way to the far right in 2024, on a calendar year basis, it was 93% combined ratio, 28 percentage points of improvement since 2017.
For those who know the business well, even if you perfectly underwrite and change everything in a portfolio, it takes a while to earn in. I mean, if you do not have every effective day coming up on January, it takes a year, then it takes a year to earn in. This is a dramatic change in terms of improvement from where we started. This is how it will compare to the industry.
This is what I just took you through, the 121 to the 93. The industry average started at 103 in that period of time and worked its way down to 96. If you take our peer average, you see they started at 99 and they improved to a 94. The upper quartile of our peers started at 96 and found their way to a 93.
Since 2017, we found our way all the way to be in the upper quartile performance of our peers. It was not the case in 2017. I hesitate to comment on this slide because it kind of speaks for itself, but I will. This is again another unprecedented because no company would have survived that 10-year period. From 2008 to 2018, the company had $33 billion of underwriting loss.
We began the re-underwriting, the repositioning, the reinsurance, which we will go into in detail. We started to bend the curve, as I said. It became a profitable portfolio. We started to become more reliable, more consistent. You can see over the last three years, we have averaged around $1.9 billion of underwriting profit. That is very interesting. It is unbelievable.
Just think about the portfolio that we had to work with in order to get there. It really took dramatic and drastic change, discipline, and execution. One of the big parts, we've talked about this, not on every earnings call, but it comes up frequently. Again, there are four numbers I want to draw your attention to here. In 2018, our gross limits were $2.7 trillion.
If you go all the way to 2024, it's $1.4 trillion. It's almost a 50% reduction. You would expect to see a dramatic falloff in premium if you were able to reduce your exposures by almost 50%. That did not happen. Yeah, we took a couple of steps back. A lot of it was reinsurance. A lot of it was repositioning the portfolio. We were at $14.8 billion. Today, we are $16.8 billion with roughly half the exposure. That has been through excellent underwriting.
Excellent underwriting is risk selection, limit deployment, attachment point, terms and conditions, and pricing. It is largely in that order. Let's talk about catastrophe for a moment. These bars represent the percentage of loss ratio and contribution towards the combined ratio from CAT from AIG. You see in 2017, it was 16%. The last four years have been very consistent, 5%, 5%, 4%, 5%.
That is planned. That is underwriting. It is reinsurance. That is the consistency we are talking about. This demonstrates the bar charts are the nominal, not inflation adjusted, but the nominal industry CAT losses for each calendar year. You can see in 2017 at $134 billion. You take the walk and we know the world has gotten more complicated in CAT, not less.
You wouldn't expect to see the ratios get better for a company unless you are AIG, unless you are going to re-underwrite, manage limits, manage reinsurance, predict volatility. What you can see here on the black dots is our percentage of the industry loss. In 2017 and in 2018, we're roughly 3% of the industry loss. If you look at 2021 through 2024, it's less than 1%.
I don't believe it's luck. As I said, I use the Thanksgiving analogy. There's no wishbone on the plate. We're not hoping for it. This has been planned. This has been executed. We've become a very consistent company that you can rely on when we're writing property. By the way, the combined ratios, even with, as we outlined, some of the reinsurance, have never been stronger.
We do not talk too much about severe losses, but it is an important part of the journey as well. How you define a severe loss is it is a single loss that is greater than $10 million. What you would see in the $628 million in 2018 would be the aggregation of all the losses greater than $10 million. We had $628 million of severe losses in 2018. That contributed 2.3% to loss ratio. Now, if you look at the re-underwriting and what the effect was, you look at over the last five years and severe losses have decreased by 70%. We are now 1% or less. That is a reflection of all the great work that we have done. I do use the analogy. If you think some of this is lucky, I say, and Golf Hole-in- Ones are lucky.
It just seems to me the more quality shots you hit, the luckier you get. There may be some, but by and large, this is performance-based. Let me talk about and transition to reinsurance. Why do I say you have to know what you're doing? Anybody can buy reinsurance.
Anybody will sell reinsurance. Brokers love to do that. You want to go buy CAT, you can buy CAT. You want to buy risk, you buy risk. Do you know what you're doing and what you're buying? I think, look, we talk a lot about it because it's an important part of our strategy. With Charlie Fry here in the front, best reinsurance buyer in the world, this is a very strategic part of our business, and it's different.
I don't think there's any place in the world where reinsurance is bought by someone of his seniority or experience, and it doesn't come right to the group CEO. I mean, it usually goes into a business or goes into the CFO. It's not strategic, and it reflects in results. How do we think about reinsurance? A lot. We also think about the details that go into it. We start with balance sheet protection.
What's the financial impact? How do you reduce volatility and control it? It becomes more predictable. What's the gross underwriting strategy? Because you need to know what you're going to put into reinsurance. First, you need to know what you're underwriting. Having that correlation is really important. Your enterprise net underwriting strategy. You have to take a long-term approach to this, and you'll see how it's evolved for us.
You can't buy it year to year and then change next year. This is all about building to reflect the portfolio, the strategy, and the execution. Now, people want me to take the slide out because sometimes it can be confusing, but I won, and we kept it in. There are a few things here that I would just look at. One is if you look at just a gross portfolio, you have obviously much wider outcomes.
We see that in the industry with companies that take significant nets. You just don't know. It is further to the right on the curve. If you buy reinsurance, it is to the left. Therefore, it means that you have more predictable results. You are more likely to understand what the profitability is. Standard deviation is tighter.
You can never eliminate the box. It reflects that you can substantially reduce tail risk by understanding your portfolio, modeling it, and coming up with a reinsurance strategy that is very thorough. Gross, you tend to have the theory, the academic exercise is you're seeding off profit. That does not take into consideration real volatility that exists in the world and exists in terms of portfolio. This is how we look at the enterprise.
It is how we actually start modeling reinsurance. Back to 2017 and looking at what reinsurance was purchased, what was the risk appetite compared to today, it is dramatically different, as you can tell. $1.5 billion net retention. That was a single occurrence, by the way. If you blew through it, there was nothing left. We have $500 million in North America commercial.
Japan, you can see the difference is, again, dramatic decrease in retention so we can control volatility. Rest of the world, they did not buy reinsurance, so there is no real comparison. We like $125 million as a retention. We do not talk a lot about property per risk. That is an individual risk loss. They could take up to $600 million net. We take $25 million net. It is just controlling that volatility.
If you said, "Reinsurance is expensive, and if you are buying more reinsurance, of course, you have less volatility. What does the cost mean?" It is a cost goods sold. I mean, I always ask. Never really get a good answer. What is the difference between what you would buy in reinsurance versus what you ought to fund net? A lot of times, people change the subject because the answer is probably not much.
I think when we look at this, it's a fully loaded cost that goes into our products. Therefore, we know what the premium is going to be and what the expected margin is. Casualty, same story, if not even, I think, more dramatic. Well north of 100 net in the U.S. historically. It's $12 million today. In international, pick the number. You don't buy reinsurance. It's whatever the largest gross limit you put out.
It was a significant exposure on an individual risk basis, but also on a portfolio basis. That's sometimes what you've seen historically at AIG, just the volatility. In international, we have a $15 million retention. Look at the different return periods. I won't spend a lot of time on this slide.
What I will do is point out what I think is really important is that when we look at return periods, so 1 in 10 is frequency. 1 in 250 and beyond is severity and more tail. We did not structure our gross underwriting or reinsurance to any one particular return period. You can see it is consistent all the way through, which is best in class. That is what you want to do. A lot of times when you re-underwrite, you eliminate the tail.
You may pick up more frequency. If you want to eliminate frequency, you end up picking up more tail. The gross reflects what the underwriting did. You can see at the 1 in 10, it was a 46% reduction in the PML, probable maximum loss. If you applied reinsurance, it was 70%.
You can look at every return period and see it's very consistent. If you go to the 1 in 250, same story. It's a 43% reduction in gross, 70% reduction in net. That goes back to the chart that everybody wanted me to delete. You end up having better net retentions. You have better predictability of volatility. You can see the substantial reduction in the portfolio during that period of time.
This is for worldwide all perils. Casualty structure for reinsurance, just because I want to show you something that I think is really important for everyone to understand. We talked about international having a $15 million net retention. We buy $85 million in excess of that. It's 100% place. John Hancock, who runs international, him and his team issue a $100 million policy. Most loss we could have is $15 million.
North America, the reason why it's 50x of 25, we don't deploy hundreds anymore. If we do, we'll buy facultative. We purchase a treaty to reflect the gross portfolio. So 50x of 25. We buy a 10x of $15 million. We have a quarter share to supplement that. That's how you get to the $12 million retention. That's the only other slide that people didn't want me to put in, but I'll explain why I like it, I think. Be patient with me on this. The 10x of 15, a lot of times insurance companies buy excessive loss to protect for vertical loss, which is a good idea.
What happens because this is not CAT where you are not going to have, I mean, in today's day and age it could, but so many different catastrophes that you need five, six, seven, or more limits. And casually, you might. If you look at this, the 10x of 15 has $320 million of aggregate available. That means there are 32 reinstatable limits. There are no APs. I mean, we pre-priced this on casualty, so it is fully embedded. If you hit it, you can hit it 32 full times and still have reinsurance available. The 50x of 25 for US, you have $525 million available. In international, you have $765 million of horizontal. You have vertical protection, and you have multiple limits that you can reinstate.
Just think that's really important because what it says up here is that extreme tail scenarios, companies tend not to buy at a return period. They buy a couple of limits. We did it in modeling at return periods. I think it's really important to look at that and stress test it. What does this all mean? I'll just give you an example.
At the 1 in 250, which is already starting to become tail, if I actually had a bigger chart or was allowed to design this the way I wanted to, I could have brought it out to 1 in 10,000. That's where it exhausts. At 1 in 250, you can just see that even if we hit that, which is starting to become extreme, we still have approximately 50% left in the international. We have 65% left in North America for the 50x of 25.
We have almost 50% left for the 10x of 15. This is built for vertical and horizontal and extreme tail. I think it's very conservative. Can't really go into it in great detail on earnings, but I wanted to show everybody this because I think it's incredibly responsible. You'll now know why we're confident in terms of large losses. If there's things that happen to go different than that's modeled, we have unbelievable reinsurance protection. Let me give an example of a business, like what's happened. North America excess casualty, we look at the total exposed limits. If you look at the green boxes, that's a net premium written. If you look at the blue bar charts, that's the net we've had for each year.
If you look at 2019 all the way out on the green line, that's the cumulative rate increase. The white line is the gross limit reduction. The conclusion is that we started off with $243 million in 2019 with a $12.5 million maximum net retention. If you go all the way to 2024, what happened? 254% cumulative rate increase, gross limits reduced by 80%. The net premium written is $474 million. Part of that is we readjusted some of the reinsurance, but most of it is through excellent underwriting, Barbara Luck and her team, driving outcomes, and the business is performing exceptionally well.
You can just see not only do we have confidence in the portfolio, the way it's been underwritten, we also have a lot of reinsurance protection and a very thoughtful strategy in terms of how we're going to position that business. Let me transition into who we are, our global businesses. Let me start in North America. Very balanced, $8.5 billion. You hear us talk a lot about, and we will today, Lexington and the excess and surplus lines.
We also have very big businesses, retail casualty, financial lines, Gladfelter, which you knew we acquired, our programs business, and also our retail property. All very strong businesses with lead underwriting capabilities and very well respected in the industry. Our distribution, 55% retail, 23% wholesale. We have these other distribution channels. Could be agents or specific programs that support Gladfelter programs and captives.
I know there is a view in terms of our large commercial, which I think is actually a positive, but 45% large commercial and 55% middle market, small commercial. We define that with clients with revenue of a billion or more, a billion or less is in the middle market. I think it is a good snapshot of the business and its balance.
Here have been the results. Really good growth. Do not forget at the beginning, we are still, I do not want to say re-underwriting, but pruning the portfolio. I think the growth probably is artificially lower than what I feel like actually the momentum is. You can look at the accident year combined ratio. I mean, memories fade, but North America commercial was the focus on every call because of the portfolio, how it performed, its combined ratio.
And of any guidance, including what I gave you today, I've been the most nervous about is saying we get below 100 combined in the business because that was hard based on where we started. We did it. We never looked back. Don Bailey and his team have done a phenomenal job. You can see 16.7 percentage points of improvement on an accident year combined ratio from 2019 - 2024.
Calendar year, similar story. We started where we started, but you see a significant improvement to 13.8% from 2019 to 2024. The performance has been just outstanding. We really have a balanced portfolio now of profitable businesses. New business. I think Don's going to take us through how he has focused on that, where we're absolutely targeting the best risk-adjusted returns and how do we actually start to turn that on. It's really important.
You can see the strong new business. You can see the contributors, Lexington property, Lexington casualty, retail property over the course of a year. The opportunities present themselves. We're prepared to take advantage of that. All right. You can see this a couple of times. The E&S market has changed dramatically. If you go back to 2018, $50 billion. If you look at the 7% above that, it represents the total non-admitted premium as a percentage of the total industry direct premium written.
The 14 represents Lexington's percentage of our overall net premium written within North America commercial. I like to think I'm pretty good at math. I can certainly count by ones. I mean, if you look 7, 8, 9, 10, it's already going to 13. I think that's pretty impressive. It's telling a story.
Those companies that have mostly agency distributions or do not really want E&S to be successful are not talking about the fundamentals. I am talking about this is growth. This is market share. There is a lot of momentum here. There is just a different wholesale market today. What does it look like in the future? 10% compound annual growth is conservative and very realistic because they have three different channels: the normal E&S, agency placement, MGAs, MGUs, delegated authority. Pick your bucket. It is all growing.
You can see it is grabbing market share. We are no exception. Our industry-leading capabilities are allowing us to deliver those returns. If you look at this, again, you measure KPIs. I like to look at new business submissions. This is unbelievable. The bar charts reflect the gross premium written from 2018 - 2024.
We were getting 30,000 submissions a year in 2018. In 2024, we got 300,000 submissions. It is 10 times the volume. No, we are not getting to all of it. When we do business reviews, the team's done an amazing job. Net premium written growth is very good. New business growth is good. We talk about it every earnings. This presents a real opportunity for us in the future.
It does show that there is a lot of change going on in the industry. Let me transition to international commercial. Very diversified, very high-quality business. It is going to have a similar story, but it did not start where North America commercial did, led by John Hancock. You can see we have a big specialty business, global specialty, Lloyd Syndicate and Talbot, and then financial lines, casualty and property round out the balance.
$8.4 billion net premium written. These businesses really demonstrate leadership in our international market and present us with really unique opportunities. Distribution is a little bit different than North America, as you can see. Global brokers of the large is at 46%. We have international brokers, domestic brokers, agents, and partnerships. There is more diversification. That is what you should expect in international. Large commercial represents 56%, middle market and small commercial 44%.
Great chart. 6% compound annual growth, 13.4 percentage points of improvement in the accident year combined ratio. It started at an underwriting profit. That is the big difference. World-class combined ratios and very good improvement on the calendar year as well. Just executing on the business and actually trying to shape it. We have done an exceptional job with that. The businesses we want to grow. Global specialty being the number one.
New business has been very consistent, very strong. You can see, as I just mentioned, global specialty has been the big driver of that. The syndicate has been a major contributor. We see opportunities in casualty. We're selective. We're disciplined, but we're prepared. We saw some growth there in new business in 2024. Global personal will have some attention in this discussion today. Very good balance.
What I will draw your attention to in 2024 is we still have travel in there because travel was still here. It is a very, very good A&H book, Auto & Home, high net worth, and it is rounded out with warranty. The A&H portfolio has always been something that has had a great reputation at AIG and one that we believe will continue. We have a growth strategy. We'll take you through that.
Geographic distribution is very balanced, Japan being such a large country. In personal, North America is really driven by high net worth. Europe and Asia-Pac round it out. It has had mediocre growth. We want to do better. The combined ratios are not where we want them to be. We will give you some guidance in terms of what we expect over the next three years on an accident year and combined ratio. It has been very steady. We like the business. We like the segments we are in. We like the countries we are in.
We have a plan to take you through as to how we are going to reshape that. Operational excellence is a core competency. Everybody knows AIG 200. To the left are some of the distinct work streams we had. This was so incredibly important. The underwriting culture needed to change.
I would say the operational culture had to change more because we really only have one. How do you build an end-to-end operational process is really important. Some of the key accomplishments, it was all about building the foundation for the future. You will see when we talk about AI that we have been able to do that. We achieved $1 billion of run rate savings with a $1.3 billion cost to achieve.
That is very tight for a project like this, particularly when so many of the investments were highly dilutive. Speaking of which, we invested $500 million to digitize our commercial underwriting platform end-to-end. The returns, of course, you are going to be able to do renewals better, new business, but it was dilutive. It needed to be done in order to shape the company for the future. We reshaped our operational infrastructure.
We upskilled talent, added, supplemented to what we had. We scaled our public cloud adoption. We went from 20% to 80% in two years. We saw an opportunity as we have across AIG in multiple cases. Like going linear is not getting us there. We have just got to transform and jump. And we did. We did an incredible job of getting that done in two years. When we did that, we eliminated 1,200 legacy applications.
That will sound boring, but when you start to get some standardization of how you are going to move to the cloud, you do not want to just forklift everything. That is expensive. We were able to eliminate a lot of the applications that were no longer fit for purpose. We modernized our data and document foundation. That feels like a throwaway line.
It's probably the most important line out of the key accomplishments because it enabled us to have single definitions for data and then to take on what we're doing today in AI. Strategic divestitures, those are usually done from a point of weakness. We did it from a point of strength. How do you shape the portfolio? How do you position the company for the future?
What do you want the company to look like in order to have outperformance? You know the strategic divestitures. Two biggest ones were Corebridge and Validus for different reasons. With that, we transferred 13,000 full-time employees, which was $1.1 billion of less cost. We eliminated $250 million of stranded costs. A lot of companies can't get the stranded costs out. We were all over it. We eliminated those costs very quickly after the divestitures had closed.
With the migration to the cloud, we needed to get out of the legacy data centers. We sold those. It was not only a $100 million proceed, but more importantly, it reduced our annual run rate by $25 million. It also enabled us from a cyber perspective and from a risk perspective to be very focused on what we're doing with the cloud and not have legacy data centers as part of our footprint. We reduced our real estate by over 40%.
Last one, which we talk a lot about now, is AIG Next. You can see we had, and you know this, we had a significant amount of costs in parent and other operations. It wasn't always clear. We needed to get to a future state operating model that we'd have a lean parent company that we would eliminate expenses, eliminate duplication.
What was not eliminated would be transferred into the business. The business would have to absorb it. The combined ratios are not going up. We did this at pace. You can see as we look to the future, we are going to hit our targets: $500 million of exit run rate savings, get to a lean parent company of $350 million, and allow us to recognize a target operating structure.
We accelerated things by creating and offering a voluntary retirement program. It just allowed us to get on with the structural changes and move the organization forward. It enabled us to have investment in future AI digital data. Let me briefly talk about AI. I am very, very excited about this. I mean, we never have enough time to really go through it.
Between me, Claude, and our panel today, we'll start touching on more detail and why there's substance behind what we're doing. We're focusing on this as an end-to-end process, not on the fringes, not just for expense savings. This is end-to-end. We want to focus on growth. That was the entire premise of our strategy. The two times or five times reflects there were two piles.
Like if you have a strategy for Gen AI that can reduce cycle time or improve revenue by two times, it goes in one bucket. If it's five times, it goes in another. Let's get on with it. Everything was documented. Everybody was involved. We needed to make sure that there was a path forward that we could execute on. We basically broke it down for simplification into five categories. Start with underwriting.
For us, Claude will call this underwriter in the loop, which means the underwriters are still making the decisions. We need to enable them. Today, there's so much time still spent gathering data, gathering financial statements. This is about why I say you give an underwriter the opportunity to identify the 125 items you'd like to have a perfect submission. What would it look like?
They do that work. We need to make sure there's consistency when you're extracting that data. Does it come from a bunch of different sources? If you ask underwriters, typically, they will start to get the same things, but they get it different ways. It makes the process very inefficient. We want consistency in data and extraction.
We want to talk about the cultural shift of how do you—you can't just overnight say, "Okay, you're not going to do five times the amount of underwriting today. Here you go. Can't wait for the outcome." You got to start to shift what happens with people, how do you train them, how do you prepare, and how do you engage with distribution. That's where we started.
You go into the data where it comes in from brokers, agents. I can't even count the number of forms that come in. You're not changing broker behavior. You get it. It's structured. It's unstructured. It's PDF. It's text. It's whatever it is. You got to be able to extract that real time to get insight, to be able to fulfill what we're looking for in those 125 pieces of information. There's AIG systems.
We've seen the account. It's a renewal. We used to have it. There's data. How do you extract it within AIG to be able to supplement what comes in through distribution of brokers and agents? Last, if it's in neither of the first two categories, how do you go to reliable sources that are approved, that have great data that will help you underwrite and fulfill those specific requests that an underwriter wants to make great decisions?
We've had all of this done. That's why Alex will be here today. We've done it with Palantir Foundry. It's an unbelievable outcome in terms of what we get. You could give underwriters—and I'm going to exaggerate for effect here—I could give them unlimited time, two, three, four weeks.
They still can't get the amount of data that I can get within two to three hours through Palantir Foundry. Okay. That is not going to just magically rub a genie bottle and end up in the underwriter's pocket. What are we doing? We have to focus on how do we actually train large language models to know what to go to extract.
Now, of the 125, we do not need to train large language models to confuse it. Can you please get the client's name? We hopefully have that. I mean, there are some data points that are already put in. The very important insightful data that needs to come to the underwriter is trained to the large language model. We use Anthropic, Claude 3.5.
Now, I thought Claude Wade, who you'll see, for the first six months, I thought he was just making up the large language model name to name it after himself. I guess it's really a model. Dario will take us through that. We started at Claude 2.0, and we are now at Claude 3.5. Just the unbelievable advancements in these large language models to be able to extract data real time from the data ingestion, the conversion into the underwriter is unbelievable.
We could also—not from an algorithm, but we could also prioritize risk characteristics like an industry group, like a geography, like the size, and then start to train the large language models to start to prioritize submissions so our underwriters are actually reviewing where we believe the best risk-adjusted returns, but also the prioritization you would see, hopefully, the bind rate go up as a result.
The data augmentation and learning with source means we can augment data. There may be a request, but there is other information that can be helpful that will come to the underwriter. I also talked about hallucinations 12, 24 months ago. That was very prevalent. Yes, it still exists. However, we can now look at the source of where the data came from. The underwriters can say, "Oh, that is very good.
It came from a rating agency or came from something that is incredibly factual, and I would have got that myself. They can actually go and check it. That is going to be a dramatic change for us. This is not—I mean, I go to WEF and say, "Can I meet with somebody that's not going to give me a pilot or something that's theoretical? Is there something that's actually really happening?" This is happening for us.
It is something that is real and something that's going to be a big part of our company going forward. We will optimize the portfolio. That turnaround was optimized in the portfolio. We already do it. Can you get it from a manual process to what we just did with our special purpose vehicle, which Charlie will go into more detail?
Can you actually do that across the portfolio more real time? The answer will be yes. We want to kind of get the data ingestion, large language models, and underwriting a little further along. The answer is absolutely yes. You will be able to allocate capital better for better risk-adjusted returns. We have a lot of examples as to where that will take place. I remember I said I would come back to this slide.
It was about the E&S market growing. It was about AIG's percentage of the E&S growing for us. What does it look like in the future? If you said, "Okay, I believe in the 10%," that means in 2025, the industry for E&S will have tripled since 2018. The submissions, 30,000, 300,000, 10 times. We have a choice. You can see what we did. We restructured.
We had a reposition. I would say we were the catalyst that drove the E&S market to change because we had a multiple distribution channel, large limits, changed the whole thing, had to flip the portfolio. Did that in 2018. Everything started to change. I think others in the industry were followers. We set the mark and were able to deliver year after year after year.
Okay, we did that. It is the strategic turnaround. Are we known? Are we back as a company that's incredibly well-respected for underwriting, for our insights on risk? The answer is yes. We start growing. New business starts flowing. Can't get to it all. If you look at the bottom dotted line, can we keep doing 10% and stay with the industry on a CAGR basis? Absolutely, we can. I mean, that would be very good by normal business standards.
I want to challenge us to think differently. I want to think about 20% plus. How do you grow 20% with what I just showed you in terms of data ingestion better, large language models, getting it to the point of sale, the underwriter, to be able to "more" and bind more over time? Let me just give you a practical example. What did it look like in 2018? In 2018, we had 30,000 new business submissions.
Buying a submit was 4%. I do not know if that is good or bad. Seems low to me. Average premium size was $260,000. And we did $300 million of new business. In 2024, we are spiking the ball. It is like 300,000 submissions. That is great. The buying of submits at two. Do I think that is—we just got to be better underwriters? Yes, but I also think we did not get to the volume.
Went down market, which was good. We thought it was more profitable. Average premium size $140,000. We did $1 billion of new business in 2024. That's very impressive. How about—again, I'm an optimist. I'm a CEO. I'm supposed to put up things that like, "Oh, if you do this, this is what happens." I think this is really conservative. The 500,000 new business submissions, again, just think of it over the next five years.
That's just staying in line with where the market's growing, the 10%. My hypothesis is, can we do 6% on a buying a submit if we adopt a new way of doing business? I certainly hope so. I mean, I'm not going 15. I'm going to 6. Not going to increase the premium size, just keep it the same. What happens? You get four times the amount of new business.
You're able to underwrite four times as much. I think when we do that, it's not going to be just adding volume. It's going to be better risks. I don't even know what happens with the market. I can give you a view, which is they're going to want to come more. I think the $500,000 will be light. We're going to be able to bind more. This is the direction we're going.
I give Lexington as an example because it is the one that is right there in front of us. There are a lot of other businesses within AIG that we're going to be rolling this out and adopting it. It's for real. This is not something that is for companies like us. I don't believe this is a choice. We're doing this. I mean, we have to do this and drive change.
Let me end with what we are doing in our top performance and some of the things I'm most proud of. Okay. We've made significant progress on capital management. You know the story. Keith will give you the exact numbers. But we've reduced share count by almost 300 million shares. We started to have confidence in our earnings more, liquidity, and dealt with so many of the capital issues that we started to increase our dividend. We're giving guidance again. In 2023, 10%. 2024, 10-plus %.
One of the things I'm most proud of is our leverage. And what we did, it took enormous discipline at a time where there's a lot of moving pieces to just continue to reduce debt. Now we've gone from $22.2 billion in 2017 to $8.7 billion and a 17% debt-to-total capital ratio. We've increased our subsidiary dividends.
The dark blue is ordinary dividends, and the green is extraordinary. It went from $350 million to $2.5 billion. Then we actually were able to get $4.1 billion with the extraordinary in 2024. Now I don't have time to go through the history of AIG or what happened and why we didn't break up the life and retirement business in the past. We talked about DTA, talked about capital.
This is a big issue as well, is that how's AIG having ordinary dividends of $350 million going to service $4 billion of obligations? The answer is you can't. There's no place to go in terms of getting more liquidity. That's for another day. You can see the discipline of going from with $1.5 billion of Corebridge, which was life retirement at the time, ordinary dividends. We're still $2 billion short.
$2 billion 150 would be specific in 2017. Discipline, retiring debt, getting more liquidity, getting a capital structure that's fit for purpose for the future. Now we have a surplus in 2024 where we generate $2 billion 750. That does not include the extraordinary dividends. The blue is our dividends from Corebridge. We are self-sufficient, generating liquidity, generating profitability, and we're all over the uses and the obligations.
If you look to 2025 through 2027, we should be generating around $3 billion of ordinary dividends. It may not be in 2025, but it's between 2025 and 2027. Our uses will be $2 billion or thereabouts, generating $1 billion of excess liquidity just based on how we run the business compared to 2017, not a chance. How are we going to drive earnings and earnings growth for AIG in the future?
Global personal, we've talked about it. There's lots of opportunities. We're going to improve our combined ratio of over 500 basis points from today over the next three years. You can see, yes, we put every category down because we believe every category can improve. Got to have more growth, improve an accidental loss ratio, reduce CAT, acquisition and expense ratio will improve.
We'll get here. I mean, a 94 is an aspiration, but it's a very realistic objective. We talked about expenses. We went from 38 to 33.2 this year. Take out travel is 32.4. We'll get below 30. I mean, we're going to have a lot of different ways in which we can do that. I ask you to give me a little bit of a benefit. In 20 years, that many of you know me, I don't miss expense targets.
I just don't, haven't, won't. This 30% will happen. We're all over it. With AIG Next, with AIG 200, let us go at the pace where we know that we're investing along the way and we're driving outcomes that are sustainable. That's the ask, but we will absolutely get there as fast as possible. M&A, we have financial flexibility. We have strategic intent. We're going to be very disciplined.
Some of the key characteristics and assessment criteria we use is, are there product capabilities that could complement where we are today? Is additional scale in a business that already has scale helpful to us? Maybe access to new geographies helps our global footprint? We'll be relentless on focusing on culture and that it's an actual fit within AIG and that it's an underwriting company. That's what we want.
There may be some adjacencies that do not exist today that we may want to explore over time. The financial targets, though, it will be accretive to EPS. It will be accretive to ROE, and we will be very thoughtful in terms of its impact to tangible book value. You will see this slide. I do not know, maybe four times. Keith has it, Q&A. Here it is. I mean, this is how we are going to build the $9.1 billion in 2024 to the $10 billion-$13 billion in 2025 to 2027.
I do not need to read all the levers, but underwriting income is going to grow, expense optimization. We have opportunities in that investment income. Interest expense, we have done a lot with debt. I would not expend a lot there, but we may get some more. The other two variables are going to be capital management and how we think about tax.
There is a lot of levers to pull. As Keith will tell you, do not try to measure the boxes because they are put up there equally, so there is no discussion around it. They are all going to contribute, and we will get there. One more note on EPS is that I think this is dramatic. I think it is exceptional. I want to point it out. If you look back to 2019 through 2023, Corebridge and Validus Re made up more than 50% over that period of our EPS.
When we made the accounting election in 2024, that we are just having EPS from AIG, it was bold. It was right. Here we are. When we look at what we did in 2024, I think it was exceptional. We gave up a lot of EPS in order to replace it with core earnings growth that is sustainable.
If you look back, we need to reposition this business to be leaner, more profitable, more consistent over time. That's the 20% CAGR. Look at what we've done. I mean, in 44% through 2022 to 2024, and then of course it's the 20% going forward. You'll do the math, but you get to 2026 is that we've replaced all of Corebridge's and Validus's earnings in basically three calendar years.
I don't know. I mean, we'll have to talk at the break. Are there other companies been able to do that and have confidence in it? We are and are really proud of what we've been able to do here. Yes, some of it's capital management, but some of it's really driving core operating performance consistently. Where we start is where we end. This is the guidance. EPS at 20% plus over the three-year period.
Our core operating ROE, 10-13%, but we're going to do the 10% in 2025. The GI expense ratio will be below 30%. The expected dividends per share in 2025 and 2026 will grow more than 10%. Thanks for your time. I'd like to welcome Keith Walsh up to the stage.
Thank you, sir. Appreciate it. He looks a little off. I made him buy a new suit. I couldn't take the brown anymore. Thank you for that, Peter. It is wonderful to be here with you all, see so many old friends, and we have a great story to tell. As Peter just outlined, AIG is not a turnaround story anymore. It is turned around. What I'm going to take you through today, we're going to go through the balance sheet in some detail and talk about the financial flexibility we've built.
We're going to continue on the balance sheet and go through our reserves and investments. Then we're going to turn to capital management and expenses, and then finally finishing with the ROE and targets. The company on the journey that we've been on has methodically built this balance sheet for strength over the last many years. We're going to walk you through that today.
Let's start with our insurance subsidiaries. We have tremendous flexibility within our insurance subs to grow. If you look at 2024, we finished with about $24 billion of premium on $27 billion of stat capital. As you know, at 89%, that ratio has come up over the last several years from 72% - 89%. We have more room to grow. What does that mean? It means we can self-fund our growth both organically and inorganically.
In other words, as we grow and add premium on, we need very little capital to add incrementally to do that. It's highly ROE accretive. What else? Within our insurance subs, our cash flows have grown dramatically. Peter has talked to you about this already. In 2020, we were able to dividend $1.3 billion from the subsidiaries. $4.1 billion in 2024. That's a 3x change.
In fact, it's a 4x change when you look at ordinary dividends. As Peter mentioned in one of the last slides, we believe a good target run rate for ordinary dividend distributions is about $3 billion on our run rate basis. That's before any extraordinary as we move out. These cash flows are a direct result of the underwriting improvements that have happened in the company. We expect these will continue to grow as our underwriting profits and stat surplus grows over time.
What does that mean? We've generated financial flexibility within our insurance companies. We've also built more debt capacity on our balance sheet over that period of time. We've paid down $18 billion of debt. More importantly, since 2020, we have simplified our capital structure with the separation of Corebridge. Additionally, we mentioned the 17% debt-to-cap ratio already. We went from probably the highest leverage ratio in our peer group to one of the lowest. This is foundational for our growth.
We have financial strength. We have flexibility. We have the ability to play offense and defense in any macro environment. Let's put this all together. I'm talking about the balance sheet. We've got financial strength. We've got flexibility. $8 billion almost of parent liquidity we ended 2024. 407% risk-based capital ratio within the U.S. We are well above targets in all of our major international entities.
We mentioned the debt-to-cap. Of course, over the last two years, we've had positive rating actions at all of our rating agency partners. Good companies have good ratings. We are committed to that. Let's pivot now to our reserve position. This is another good news story. Frankly, it's one that wasn't such a great story six years ago when the new management team came in and started changing some of the underwriting.
We've seen the reserves follow. We've gone from deficient to what we feel is a very healthy level that we're proud of. What are some of the principles? We have $40.1 billion of net reserves on our balance sheet. One of the biggest principles, and you hear us talk about this all the time, is recognize bad news quickly and allow favorable trends to develop. That is a core principle.
Additionally, developing conservative initial loss picks, right? Watching them develop over time as well. We need strong and frequent triangulation between reserving, actuarial, underwriting, and claims, right? That is a relationship of challenge and of collaboration. It's critical to what we do. We do deep dive reserves on an annual basis as well as monthly and quarterly.
As we touched on on the last quarter call, we have a holistic way of looking at our reserves, and we will react every quarter. We're not going to just highlight on the annual reserve reviews. We'll react to news as we see it. Let's look at this in practice. If you look over the last 10 years, you can see the underwriting improvement has translated to our reserve development. From 2015 - 2018, we took over $10 billion of reserve hits.
That was an 11% hit to our book value per share on an adjusted basis. Over the last six years, which coincides with the improvement in our underwriting, you see six consecutive years of favorable development, which is a 3% impact cumulatively to our book value on an adjusted basis. Now, we do not rely on this. We do not budget for this, right? It is not a major part of our earnings.
As you saw in some of the prior slides, we have about $2 billion of underwriting income over the last several years. This is not a large percentage. I want to touch on a couple of pieces of analysis that I know get some attention. What we want to do here is we want to look at our U.S. long-term commercial Schedule P data over the last 10 years.
What do we mean by US long tail? It really is other liability occurrence. It is workers' compensation, and it is commercial auto. Three of the lines that get a lot of attention. If you look at this, this is our initial loss picks versus the industry. As you can see from some of the earlier years, we were well above the industry, and we should have been well above the industry.
You see some of the actions that we took. On accident years 2016 to 2019, back in 2019, we took $1.2 billion of reserve strengthening on those years. That is $900 million on our other liability occurrence line and $265 million on commercial auto. We took our loss picks up significantly higher than the rest of the industry and higher from where we were.
Additionally, you can see as the lines start to converge over more recent accident years, that's from our re-underwriting as our book has become very profitable, consistent with the rest of the industry. Now, we didn't take those picks down in those prior years, and we moved earlier than others. 80% of these actions were done before 2020. We feel really good about our relative positioning.
Let's look at the same analysis drilling in specifically on the left to other liability occurrence line and on the right commercial auto. You see the same relationship here. Additionally, we got cumulative rate on these long tail lines of 82% since 2019 and 108% excluding workers' comp. Additionally, we put in loss cost trends of 10% or greater on all lines on all of our workers' comp business post-2022. We were at a significant number before that.
Let's look at another analysis at IBNR. What we're doing here is we're going back to 2019, which represents about the six years, five to six years of the re-underwriting strategy that's taken place. What you could see from this chart, and these are 10-year rolling accident years. For example, in 2024, that's our 2024 year plus the nine preceding years before that, 2023, 2023, nine years, etc., as you go back.
What you can see is that our IBNR as a percent of total reserves went from 68% to 80% over that time, a 12-point differential. You can see for the industry a 4-point increase. We think this is a prudent metric. Once again, drilling down into those specific lines of business, you can see the relationships even more pronounced.
We have a 14-point increase over that period of time in IBNR as a percent of total. The industry is at 5%. In other words, we feel really good about where we went on our reserves. We were deficient to strength. We reacted early, took those loss picks up, re-underwrote. We feel like we have a very healthy position. Let's turn to investments. We have a high-quality portfolio, about $90 billion.
Frankly, over the last five years, we've been defensively positioned, really all about maintaining capital, liquidity, right? Just more defensive. We have an opportunity as the underwriting improvement has taken hold to be a bit more strategic around the portfolio. A-plus rated credit, 3.8 years duration of assets, which matches our liabilities. Let's dig in a bit more. There are two pieces I want to talk to you about.
The first would be around our core general insurance portfolio, which is $77 billion of that almost $90 billion. The second is our other ops portfolio, which sits outside of core GI. When you look at the net investment income driven by both of these pieces, it's about $3.5 billion.
$3.1 billion in 2024 was driven by the core GI piece. Other ops drove $420 million, $424 million to be exact. What I want to talk to you just first about other ops because we have a lot of moving pieces here. What does that make up that $12 billion? It's primarily two things. It's about $8 billion, which is our parent liquidity number we showed you earlier. And then there's about $4 billion of our remaining stake in Corebridge.
Now, the $420 million that sits in other ops net investment income is going to materially decline over the next couple of years for three reasons. Number one, interest rates are lower than they were a year ago. Number two, parent liquidity, we use that money for our share repurchase. As we buy back stock, that number will naturally come down. Number three, as we continue to sell down Corebridge, we will receive less dividend income. That is where it sits. That number is going down, okay? Let's move on to the $77 billion because this is really where our core strategies take hold. That is the $3.1 billion. We are not immune to short-term rates in the core portfolio as well.
What I would tell you is that for 2025, when you take all that together, we won't get a lot of growth in our net investment income because of some of those dynamics we just talked about. For 2026 and 2027, we expect our strategies to take hold and to grow our net investment income. We'll be an important contributor on our ROE journey.
There are really four things we're looking to do. Number one, our invested asset base should start to grow again, right? As we've divested businesses over the years, we have not grown the invested asset base that much. That will start to change. The second thing, we have some incremental opportunities in parts of our portfolio overseas, specifically in lower-yielding places such as Europe and Japan to get a little more yield. Third, we will continue to invest our runoff.
As our portfolio matures, we reinvest at higher new money yields, roughly 100 to 125 basis points above the current level. Lastly, increasing allocations to private credit. We are talking about modest increases here. We are currently about 8% of the GI portfolio. We expect to take that to about 12%-15% over the next several years. Of course, you get higher yields than you do on public equivalents, so about 150 basis points.
As that takes hold, that will help us. In other words, a very liquid, strong portfolio, and we believe we are taking actions to position it for upside. Let's turn to capital management. This is the fun stuff. What do we do with all the money, right? This is a snapshot. Peter's touched on some of these topics already. We have spent $45 billion on basically three categories since 2019, right?
What is that? $21 billion on leverage reduction, $18 billion on share repurchase, and a little over $6 billion on shareholder dividends. One of the key messages I want you to walk away from today is that every dollar we earn, we're done divesting, we're happy with our portfolio. Every dollar we earn on a go-forward basis is for productive use in growing our business, investing in our business, and returning capital to shareholders. That's it. Let's drill in a bit more to share repurchase.
Peter mentioned this in detail in one of those last slides that you saw. We have used share repurchase as a tool to right-size our equity base as well as rebuild our earnings per share power within the company. So a little bit of information here. There's a few moving pieces. This morning, we announced a new $7.5 billion share repurchase authorization. That replaces the existing authorization.
Just to be clear, $7.5 billion from today going forward, okay? We anticipate we'll buy back roughly between $5-$6 billion within 2025. Through the first quarter, we've repurchased $2 billion, a little over $2 billion of stock. An incremental $3-$4 billion over the next three quarters of 2025. Now, I want to put your attention on the right side. You've seen some of these numbers already.
We believe we have on a target run rate basis, once we're done with the Corebridge sell down, and just as a reminder, we still have $4 billion of liquidity that we can monetize from Corebridge. We've committed to hold 9.9% through December 2026. We have an ability to get more monetization and to do more buyback. On a target run rate basis, we're driving $3 billion of ordinary dividends.
Peter mentioned in one of his slides, we have $2 billion of uses. What are those? Roughly a billion dollar plus shareholder dividend, corporate expense, and interest expense, roughly $2 billion. That means our share repurchase on a run rate basis without any asset sales, meaning Corebridge, is a billion or less on a go-forward basis. Just want to make that very clear.
Let's move to dividends. Peter mentioned this as well. We grew our dividend 10% plus the last two years after six straight years of stagnation before that. This is an important metric. Dividend increases are the sign of a healthy organization, a healthy company. We believe in it. We think it's important. We've committed to 10% plus for the next two years, right? Why are dividends important?
Some of our long-term shareholders have actually published research on this where we see over long stretches of time, reinvestment is as much as 40% of long-term TSR, dividend reinvestment, right, over the last 20 years. It is even higher if you go back longer than that. We are believers in dividends. It is the sign of a strong company, a growing company, and we are committed to being a good dividend payer as we go forward. We are going to go back to the same slide again.
Peter did this. I love this slide on the expenses. I want to talk a little bit about some of this. This is adjusted for the asset sales we have done, meaning Validus, crop, etc. When you look at the reduction from $38.3 billion to $32.4 billion, that is a real almost billion-dollar takeout of cost.
The other thing we've done to try and make this more apples to apples is, as we've said to you in the past, we believe $350 million is a good number for parent expense. What we've done is taken everything over $350 million and reloaded it back into those years in General Insurance to give you more of an apples to apples look about the real change in our expense ratio. We're confident we're going to drive that below 30%.
Why? A couple of reasons. Number one, Peter said it, okay? That'll be part of it. Secondly, and more importantly, we have real opportunity. AIG Next will continue to earn in over the next couple of years, and we'll get benefits from that. More importantly, Peter mentioned this, that we have stranded costs that we had taken out.
Stranded cost is the worst two words for any CFO to hear. It is really hard to take costs out when a business isn't growing as much. As we are now, our growth strategies are taking hold. We will get normal operating leverage in this ratio that will benefit us. It's as simple as making sure our expenses don't grow as fast as our top line. I'm going to repeat this slide too because it's a good one.
I know you guys care about it. The ROE. We are laser-focused on the ROE. We know what this means. We know 10% is a minimum number, and we've committed to getting there in 2025, 10% plus. We expect to improve within that range over the subsequent two years. We've got multiple levers and multiple paths to get there. In summary, here's the targets again.
We think these are good targets. These are tough targets, but we're committed to achieving them. 20% earnings growth, 2025, 2026, 2027. Obviously not exactly the same. It's a K-curve. Growing the ROE in that 10-13% range, taking our expenses below 30%. That's an important metric. And growing our dividends per share, double digits in 2025 and 2026. Growth on growth on growth. We're just really excited to be here with you today. It's a lot to take in. There's a lot that's going on. We've got a great story, right?
And we're just getting started. Thank you very much. I would now like to welcome back to the stage our Chairman and CEO, Peter Zaffino. We're going to do some Q&A. I would also ask Quentin, we'll hand the microphone out if you could raise your hand. And then when he hands you the microphone, if you could stand up, ask your question, and just state who you are and where you're from. We'd appreciate it.
All right. Thanks. In the interest of time, we're just going to ask everybody to limit themselves to one question and one follow-up. I'm Quentin McMillan, the head of investor relations for AIG. And we'll take the first question from Elyse Greenspan from Wells Fargo.
Hi, thanks, Elyse Greenspan, Wells Fargo. The first question I have is on the expense ratio. So a lot of improvement. I think that was one of the last few slides, Keith. And obviously, you talked about expense leverage. There's also some from AIG Next. Could you just spend a little bit more time?
There's obviously been through the years, we've seen AIG at a very high expense ratio, and it seems like it's a big driver of the ROE improvement. Can you help us just go through the numbers in a little bit more detail? Over the three-year plan, is the 30% a 2027 target? Maybe sooner? How should we think about the timeframe as well?
Thanks, Elyse. We're not going to itemize the target, and we're not going to give you by year what it's going to go. We're moving in that direction. We commit to getting below 30% over that period of time. One of the things is we have about $250 million of AIG Next that will continue to earn in. Probably about two-thirds of that, let's call it in 2025, and the other one-third in 2026. That's an important component.
I can't stress enough the operating leverage piece, which we talked about. That will really just give you, just model it out. Just say if our top lines grow and whatever you think our premiums are and grow the expense base at a percentage of that, you'll start to see some of that come off. I think we've given you some really good parameters to go off there.
Can I add to that, Keith, a second? It's like premium leverage, AIG Next earn in. We have more efficiencies to gain from how we're structuring our business. There's better end-to-end process opportunities outside. Forget about AI for a second, just in terms of how we've digitized some of our businesses.
If you look at the discipline that exists within the business today, to be able to absorb some of the costs that came from other operations and parent, they created that room. We are constantly finding ways in which we can improve the expense ratio. I think some of the big countries, as you start to dig in, like we are weaving the company together is not a way to describe a business strategy. It is about weaving it together. There is a lot of duplication. There are still inefficiencies that exist. He is always the one that has to be cautious. I would be disappointed if it was 2027. This is not a hockey stick aspiration. We gave the range. I would expect us to get there earlier.
My follow-up, you guys alluded, pointed to excess liquidity generation of $1 billion per year. Does that bake in projeced earnings? In terms of buybacks beyond this year, is that $1 billion the base case buyback, ignoring, right, the sell down to when you go nine nine to potentially none of corporate over a couple of years, just base case is kind of $1 billion repurchase?
Go ahead, Keith. Yeah.
Yeah. I mean, that's what we were trying to, again, the $1 billion is without any other asset sales. As you know, we have a $4 billion position still in corporate. That's just a run rate you should be thinking about for us. Obviously, that could grow as we grow going forward. I did not quite hear your first question.
I was trying to understand if the deployable capital, that $1 billion, does that include projected earnings growth within the business?
It is based on our current view of ordinary dividends, right? As our underwriting income grows, as I mentioned, that number can grow over time as our stat surplus grows.
Thank you.
We'll take our next question from Mike Zaremski from BMO.
Hi, thanks. Mike Zaremski from BMO. One question on capital, maybe probably for Keith. Should we, over the last year or so, the RBC ratio has come down? I'm not sure about the J gap or what are the Japanese ratios. Maybe you could provide some color there. Do you expect to be able to continue to bring down the RBC ratios or the regulatory capital ratios? If so, would extraordinary dividends be in the picture as well in the coming years?
Want me to grab that?
Yeah.
Yeah. The RBC is at a very comfortable level. Yes, we have taken it down because it was way above any normal standard within the industry. I think we can take it down a little bit more. I'm not sure we will because I think that there's opportunities to grow into that. We're at a very comfortable place now with RBCs. I do believe that there are still capital efficiencies that exist, as you just mentioned with Japan and some of our international structures. I mean, they've been there historic, their legacy.
There's opportunities really across the world to be able to consolidate some of the capital structures. I would expect us to get capital out over time, meaning over the three-year period. I wouldn't expect us to be looking to give targets on RBC that are kind of below where we're running as we exit 2024.
Got it. My follow-up's on top line growth. I don't want to steal any thunder from the AI panel that's coming up. The opportunities for growth that you're talking about, are these kind of like you flip the switch on once the technologies are in place and you can start growing into meaningfully? Or does it also matter kind of how the operating environment is on a competitive basis? Just kind of curious if you think the competitive environment is directionally easing or still staying just as competitive on the P&C side.
Yeah. A few things. One is I think we can grow at or above where we've grown apps and what we're doing with AI. Again, if the market, we're an underwriting company. We're going to drive profitability. We're not going to drive top line, but still think that there are very good opportunities that exist in the market with our specialty businesses or actually what we've done in many parts of our global property have run exceptional combined ratios. You have to think of back to this cost of goods sold analysis is that in many ways, the reinsurance costs on a risk-adjusted basis are going down more than what actually the primary.
You have to think about that in terms of combined ratio expectations. I think we can grow, maintain profitability, continue to maintain discipline. Say if you keep that constant, I would expect a little bit more growth. When we are starting to talk about AI, it's hard to give a timeframe. I think you're going to hear that from Dario and Alex today.
They're more confident in a three to five-year period than they are next year. We want to make sure that we're investing to be able to capture that three to five year. Do we have to wait then? Probably not. I think that that will become aligned with risk appetite, underwriting, and actually accentuate on a risk-adjusted basis more profitability, not less. What that growth looks like over time, we won't know. I think it's going to be much greater than what we would expect in a normal environment like we are today.
Take our next question from Brian Meredith from UBS.
Yeah. Thank you. First one, just dive a little bit more into operating leverage. You've highlighted a bunch of different things you're doing to try to get growth. That 13% return on equity, where are you kind of thinking the operating leverage is going to be at the company at 13%? And where do you think a comfortable good level to be at is for the general insurance business, so premium to capital levels?
Keith, you want to?
Yeah. I think we can get to one time. I mean, obviously, it's business mix dependent. You see a lot of shorter-tail companies can run much higher than that. It's not a perfect metric. There's a lot that goes into that. It's a simplistic metric just to compare across companies. But I think if you see some of the longer-tail companies, you can get to one time. And I think that we'd be very comfortable in that range. And you clearly see we have a lot of excess capital within our insurance subs or well capitalized within our insurance subs.
Is it, Brian. You did not ask this question, but somebody else will or somebody will want to. For M&A, if we look at M&A, we think we are going to be incredibly disciplined. We know we will be. We have a criteria, think there is going to be opportunities, and expect over a period of time. Today is a moment in time, but over a period of time, I think we will be able to acquire something.
It will be additive to earnings and accretive to ROE, and we move forward. If we do not, because with our discipline, it does not meet our criteria over a period of time not to be defined, and we still have excess capital, we will return that.
I mean, we are not going to hold on to capital forever if we have too much capital relative to the size of business we are. That is not our preference. I don't think it'll be the base case. Over time, if that happens, we'll lower the equity. That'll obviously boost ROE. It won't be for that reason. It'll just be that we don't want to hold on to excess capital for an indefinite period of time.
That actually was going to be my follow-up. Maybe I had a hunch. Great. Maybe going, adding on to that a little bit, Peter, maybe you can talk a little bit about what areas do you think M&A makes sense from a business perspective? You talked about scale, culture, competencies. You don't have small commercial. Is that an area, E&S, that kind of makes sense to y'all? Other areas that kind of make sense? How do we think about M&A and kind of areas you want to potentially add to?
Look, it's harder in an insurance company than it is in a broker. In a broker, you decide where you want to go and you find assets, and then you try to embrace them into a company and a system. Have to make sure the criteria I said was like third or fourth is really important, which is good underwriting company. Good underwriting culture means it's probably got a good balance sheet.
We do not want to have to drive company to turn it around. Still, it's a big world. I mean, there's a lot of opportunities for product enhancement on what we already have. You mentioned small commercial. I do not know that we have to have it. You saw the breakout of North America and international. We already have a very good small commercial and medium portfolio. It would be good to add to that.
I mean, I think it would be a good balance. There's a lot of places in the world. I would lean more towards international, not to say we wouldn't do something in the U.S. It has to be something that would not require a lot of integration, very committed to the financial metrics, product enhancement, geographic enhancement, or something. I don't know that we need to add something to Lexington.
We could. Where we already have scale that will accelerate our path is very interesting. The other one, not to make everything about AI, is that companies that don't have scale or expertise are not going to be able to do it. They're just not. It's just going to take too much time to play catch-up. If there's opportunities in the future that a company can come in, we can integrate it and adopt the work that we've done, there could be significant accelerated growth. That's going to be something we look at as well
Next question from Wes Carmichael from Autonomous.
Hi, thanks. Wes Carmichael from Autonomous Research. On the 20% EPS CAGR, I know we can't measure the boxes there. But if we think about 2025, we've had the California fires early. Should we expect that to be a little more back and loaded 26 and 27 when you get maybe above 20% with expenses coming online as well? I wouldn't think of it that way. I mean, I don't know what's going to happen. It's March. We've got a long way to go in the year. The wildfires for us will not be an inhibitor in terms of driving the EPS expectations that we've outlined. I wouldn't think of it that way.
Thanks. I understand most of the leverage is on ROE. I guess the one on tax efficiencies, is there anything incremental? I realize maybe you're earning a little more and maybe utilizing the DTA a little bit more, but anything else there?
No, I wouldn't. It is a lot of incremental. There are a lot of things we can do better. I think as we've turned the underwriting strategy around, we have an opportunity, I think, with the tax strategy to just do better blocking and tackling. The DTA is certainly a part of that. But yeah, I'm not going to go much beyond that.
Thanks.
We'll have Q&A later, but we have time for one more question. Andrew Kligerman from TD Cowan.
Great. Sounds very impressive the work you're doing with Palantir and Anthropic at Lexington. You talked about 20% growth. My question for you is, versus your peer group, are they doing similar things? If they are, how can you be confident you'll grow?
The short answer is I don't know what they're doing. I don't think they are. I mean, when they go on to earnings calls or my just general observations is there's 300 instances, it's operations. It all sounds all interesting. Nobody that I'm aware of is doing end-to-end in terms of what we're doing with AI. I'm not particularly concerned, honestly, with what they're doing. I just know the path that we're on. I'm convinced it's the right path. I've seen some of the things that we've done early days. This is not hypothetical. It's real.
This is the way we're going to drive scaling our business. I would expect that's the environment we're going to be in in the future. I don't believe we'll be by ourselves, but we're going to be a trailblazer in terms of driving industry standards.
Keith, you talked a little bit about how actuarial and claims and underwriting talks to one another. Is there something structural at AIG that kind of institutionalizes that, or is it just more of them talking to one another?
I think it's culturally, as the re-underwriting took place, it was, I think, as Peter said, it had to happen. I think that accountability from the leadership that came in, the way we drive it, internal and external reviews, there's a lot of robust engagement back and forth. It is more, I think, the necessity out of it and culturally, the way the underwriting changed, and that has been part of the reserving process as well.
Thanks very much.
I think Quentin was supposed to say we are going to take a 15-minute break.
Take a 15-minute break and be right back.
There we go. There we go.
Ladies and gentlemen, we will now have a 15-minute break. Thank you. Ladies and gentlemen, AIG Investor Day 2025 will resume in five minutes. Ladies and gentlemen, AIG Investor Day 2025 is about to resume. Please prepare to join us.
Shit, they are already starting; they are already ticking down. All right, welcome back. Hope the break was great. Okay, we have got a great panel discussion today with three of our business leaders. They need no introduction, but I will do that just in case.
John Hancock, who is the CEO of our International Global Commercial and our Global Personal Insurance. Don Bailey, CEO of North America. Charlie Fry, who's Global Head of Reinsurance and Portfolio Optimization. Guys, looking forward to it. John, let me start with you. We talk a lot about our global specialty business. It's performed exceptionally well. I talked about that earlier. Why is the specialty business unique and really valuable to AIG, and why do you consider us an industry leader?
Okay, thanks, Peter. It's great to be here with everybody. First off, global specialty. Everybody defines specialty in their own way. I think to be clear, specialty to us, it's marine, it's energy, it's aviation, and it's our credit business. You can see from this slide, we've got scale and we've got fantastic performance in every single one of those businesses.
Yeah, and we've got a very, very big specialty business, $5.6 billion of premiums. You're right, I mean, we're big, but yeah, I think we're the leaders. I think we're the best. I mean, why do I think that? I think start with financial performance. Combined ratios around 76% average for the last five years. Whilst we've been growing our business 40%, yeah, why else?
Our customers like us. We renew more than 90% of this business. We're writing more than $700 million of new business every year. This is a really, really strong business. That strength, that size, that gives us influence. We're an influential player in this marketplace, which means now this is complex business. This is business where you need to know the industries and need to know your clients. We are influential. We're the standard setters. We're the rate setters.
We're not just the followers here. That makes a big difference to this business. It's something we've built up over decades. It's a real differentiator for us that, you know, we've spent decades building our reputation, our portfolio. It's impossible to replicate that overnight, either the portfolio or the expertise. It's also a place, you know, we're big, we're great at this.
There's also lots and lots of opportunities. I talk a lot about marine and energy, where we have clear leadership positions. Those are growing sectors, growing industry, lots of room to keep growing all over the world. Beyond that, you know, on aviation and credit as well, all the specialty lines, we have got huge expertise around all of those. It's recognized. Because of that expertise, we attract the best talent in the industry.
Ally that with, sorry, those financial results. That gives us, with that expertise we've got, that gives us the confidence to keep growing this business. Those combined ratios are not a one-off. Those are five-year averages. We do that year in, year out. It's a terrific business with lots more to come.
This will feel like a throwaway line, but I think it's really important is maybe just talk about the structure because, like, we run global specialty as a global business. It's not, like, divided by territories. You know, Gordon runs the global business.
Yeah. We do, we run it globally. There's good reason for that. That expertise, that capability, if you try to build that everywhere, that's crazy. You can't build that depth of expertise everywhere. What you can do is do it centrally and then deploy it through our hubs, which is what we do. We're headquartered in London, which is the heart of the specialty market.
We use hubs strategically placed all over the world so that we can use that expertise, deploy it locally close to our customers where we've got big knowledge. Yeah.
That's great. I'll come back to you in a second. I'm going to go to Don here. Similar, Don, we talk a lot about the Lexington, E&S growth. Again, that's something else I, you know, mentioned in my prepared remarks. We have other businesses. You know, can you share a little bit about our larger commercial businesses, how important they are, how they performed, and their relevance?
Yeah. All of the work you highlighted in your early, you know, comments, Peter, have delivered us in North America just a large scale, very healthy, highly diversified portfolio. We would say it's a portfolio that gives us great competitive advantage in the marketplace. We can be very thoughtful about where and how we're growing our business.
There's three large commercial retail businesses that we operate in. We've got casualty, we've got financial lines, and we've got property. The casualty business, you made a few comments on it earlier. It's a very healthy, it's a very resilient business that we've got. Through all of these, our ability to lead and access is another strategic and competitive advantage.
The casualty portfolio is, it's strong, it's resilient, and it's such that we're able to be on the offensive in the marketplace right now with a lot of the disruption that's playing out in the market, particularly in excess. We can be very thoughtful about the opportunities that we're pursuing.
In the financial line space, very much our brand in North America, you think about financial lines, probably D&O in particular when you think about us. That D&O is highly diversified. Again, it's primary in excess. We've been able to be very effective in strategically deploying our limits and our attachment points in the financial line space to drive profitable growth.
In property, which is in many ways a very interesting conversation right now as well with all of the rate swings that we're seeing and the risk dynamics that are playing out, we've been very nimble with our property capacity and be able to deploy that in ways that drives profitability for us in both retail and in wholesale, regardless of the market.
The opportunity, Peter, if I could just, you know, add this on to the commentary, is about how we sell and distribute those assets even more effectively. Like, that's the name of the game for us. I spent, as many of you probably know, a lot of my career on the broker side, and I got to see a lot of carrier distribution models and face off with those. I ran global sales for Marsh. I was head of Wells North America.
I was in those markets dealing with those carriers every day. Overall, I would say those distribution models were fairly low value. They tended to be very internally focused, very product focused, very geography based, and did not drive a lot of value for me as a broker or my clients. When I got to AIG almost two and a half years ago, I spent a lot of time just thinking about the distribution model as an asset to the organization.
We have created a unique model for us where everybody in AIG distribution is part of a national broker team or a national client team, where the focus and intensity is on the external stakeholder, the broker or the client. It works for us because we have got density in our brokers from a retail and a wholesale standpoint.
In retail, 83% of our premium volume is with 10 brokers. It allows us to be very intentional about the growth that we're driving and the alignment that we're driving with them. Our share of wallet with those top 10 is only 2%-7%. We can double the size of our business within each of those. Alignment is driven with this model. Deployment of target account programs is driven. We're able to much more thoughtfully use data to drive, you know, clear and unprecedented opportunity. The model works. You can see it in the new business. You can see it in the top line. We're excited about where we go forward.
Did you like your boss at Marsh?
I was, I learned a lot.
Good answer. I think your ability to have fine-tuned this distribution approach, I mean, again, there's things that sound very simple, but companies tend to be unorganized. Having an organized approach to risk by using data, I mean, like you've seen it in the numbers, but I mean, do you feel like you have more momentum now as you start to execute on this?
Yeah, we have in the data part that you just referenced there, Peter, has been a huge part of it. We spent the last couple of years again working with Claude and his team in building broker and client insights that we just never had before that I can just at the tap of an iPad see exactly what the trading relationship is with the broker. That accountability drives value. Yeah, totally.
Thank you, Don. Charlie, I've talked a lot about property cap from 2017 - 2025. For those who don't know, Charlie was the first hire that I made at AIG. And he thanks me all the time for that. Can you just take us through, you know, I outlined the differences in structures, like in philosophy, but what are really the fundamental differences from a reinsurance structure in 2017 to today? Because we've talked about that long-term journey, but just maybe you can connect the dots, you know, for our audience today.
Yeah, sure. I think the starting point, we're talking a period of time where there's been $1.1 trillion of natural catastrophe losses to the industry. You know, it's certainly a significant level of activity. You in your remarks earlier, you talked about the philosophy we have in every decision we make when it comes to purchasing reinsurance.
You go up to 17, when we arrived here, there was none. If I was going to describe the strategy, it was, it was sporadic, it was opportunistic, it was decentralized. Essentially there was not one. There was nothing like the in-depth data and analytics that we would utilize to be able to make informed decision-making. Now, I think if you compare the 2017 cat structure to our one today in 2025, it is the epitome of having a lack of that strategy.
We had very high retentions. We single-shot, which you mentioned. Therefore one event and one event only. In some places we did not buy at all. If you look at North America as an example, the company attached at $1.5 billion, had one limit, one limit only.
Now, for context, that was a year there were 17 named storms in hurricane season. 10 of those were catastrophes. 6 of those were major catastrophes. You think 3 of those made landfall in the United States within the space of a month. We saw in the earlier numbers that it was a heavy cat year for AIG, but it would not have taken much and it could potentially have been significantly worse. You then turn to Japan.
We have a sizable business that is under John in Japan. What was happening in 2017, for me, it was the epitome of the problems with decentralized buying. We had a personal lines business and a commercial lines business. They rolled up into different people. They bought separately. Of course, the businesses are subject to the same perils. We end up with two retentions.
You only have to roll forward to 2019. Two of the four largest typhoons in history in terms of losses hit Japan. It could have been a very, very different story for AIG if we perpetuated that buying philosophy. You touched on it. I mean, we got this incredible footprint and it seems surprising to me that the decision was made not to buy internationally. You do not need to go back too far in time to think through to double-digit billion-dollar events in terms of tide floods, Christchurch earthquake.
That takes you forward to 2025. We have low appropriate retentions across our business with reinstatable limit. Japan, North America commercial is $500 million. Japan, $200 million. That is one retention to be clear. Of course, we buy for the rest of the world and you have talked about that.
There is another important part of our structure, which is our aggregate cover. I do not think our aggregate is comparable to when you hear others talk about aggregates in the reinsurance market. A lot of discussion around aggregates, what they really mean is high attaching, so dealing with a series of extreme tail events. Whereas ours is much more focused around frequency.
I think there was a question earlier about the wildfires, obviously tragic event, but within our North American attachment for the aggregate, there is a sublimit for secondary perils of $450 million. You have to assume, given what has happened, a sizable amount of that aggregate is already eroded. We are really happy with where we are today.
Does it aggravate you as much as it aggravates me, like, in terms of how our aggregates compare to others? I mean, like, it's like nothing else in the industry. I just think one more click down.
Yeah, I mean, it's like, you know, it's sort of American, English, and English, English. We both talk about football and there's some comparisons. Fundamentally, they're two completely different games. Look, when we think about our aggregate, we don't think about it as standalone.
We think about it as part of our overall buying strategy. If you think what we've got, we've got really strong protection in first event, second event, third event. That's how we go about buying rather than, you know, let's see what happens towards the end of the year. Put it this way, I'm very happy where we are.
Yeah, me too. Let's transition to the SPV that we just launched and announced. Really innovative, you know, something we've been working on for quite some time. Can you provide a little bit more detail? What is it? What's its purpose and why we think it's a foundation for the future?
Yeah, absolutely. Look, we're extremely proud of what we've done here. We've collaborated with Blackstone in what we believe to be a truly innovative, actually first of its kind, some of the features of this vehicle. What we've essentially done is we created a brand new reinsurer.
But it's a reinsurer that's exclusive to AIG. Now, again, a little bit of context, it's the second largest new syndicate in the history of Lloyd's. And it's one of the most sizable capital raises for reinsurance since the class of 2005 and 2006 and KLW. And it really involved a lot of strong collaboration across all of our business.
Now, if you think through, as I said, it's long-term exclusive reinsurance to AIG backed by absolutely first-rate capital and a first-rate partner. It also provides us with access to some fee income, which is highly accretive in terms of capital base.
You then go a little bit further and you think for them, they get access to the vast majority of the treaties, a sizable share of our outwards reinsurance, which of course gives them access to the platform that John and Don run that we talked about, which is, I think it's an incredible platform in terms of our ability to originate risk. I think that's what's so exciting for me about the SPV. It's fantastic what we've achieved, but it's where we can go, which I really feel excited about.
This ability to originate risk and in actual fact match the risk and optimize portfolios to the most efficient capital available, of course, being our balance sheet, but also third-party capital. We retain the balance within our own business, but also have the opportunity to generate, you know, sizable fee income into the future. I think it's really, really exciting.
Is that 30,000 a typo or is that real?
I mean, it's in terms of modeling. It is. There's an astonishing amount of work that goes into it. I mean, that's why I mean, yeah. Of course, it's, it's, why is it take, it's foundational, right? What we're looking at, we're looking at the entire portfolio of AIG. We're talking, you highlighted and the guys have highlighted the amount of change the company's been through.
It is not just looking at the data, it is looking at the trends, applying macro trends in terms of what we think inflation will happen, you know, climate change, social inflation, but also looking at the underwriting changes we have both made but plan to make into the future. The good news is with that foundational modeling done, it gives us, we do not need to do it again repeatedly to continue to build.
But I think what I wanted to draw, Charlie, which is exactly what you just said, is that what I talked about with AI before, the work is done on the portfolio optimization. We need to update it. There are different factors, like you said, it is an ongoing model, but we are not building something else. The SPV was a portion of that portfolio work, but the foundational work has been done for us to be able to look at.
Listen, 100%. I think we're going to, you know, you've talked about it in the next session coming up. You use the word turbocharge for the business. It turbocharges our ability to develop these structures as well. That's right. Super exciting.
It is very exciting. Thank you. John, let's talk about Lloyd's talent. You personally, AIG, have a very special relationship with Lloyd's. Why is having a syndicate matter? Why does it give you advantages in underwriting? Maybe you can just bring it back in terms of, from your perspective, having had very senior roles in Lloyd's, like why we're performing really well.
Okay. Yeah. I mean, it is a really important complement to our other distribution and our other access to risk. We do have a special relationship. Me personally, lots of you will know before AIG, I was the Performance Management Director at Lloyd's. That gave me a, you know, that effectively means I run underwriting across Lloyd's. That gave me a real rare insight into the market.
I think if I may just step back, I know some people will know this, but I'm not sure everyone knows, but what is Lloyd's? Firstly, it's not an insurance company. I think that's really important to understand. It's a marketplace. It's a $72 billion marketplace. It's made up of 55 managing agents, about 100 syndicates who all trade with, compete against each other, but are also held together with a mutual interest through the Central Guarantee Fund. That's what makes the market really unique. At the middle, you've got the Society of Lloyd's.
It regulates the market, but more than that, it sets the standards. It sets the underwriting discipline. It upholds the reputation as well. It is a real glue that holds the thing together. We have a really strong position at Lloyd's through Talbot, which is an outstanding managing agency there. Currently three syndicates, you know, 2019 and 2478. 2478, the SPV that Charlie's just talked about, gives us the opportunity to deploy our capital differently or manage third-party capital to support the AIG portfolio.
That is a terrific benefit for us. Syndicate 1183, which is our trading syndicate at Lloyd's, is one of the leading syndicates at Lloyd's. You know, I said earlier, you know, I do have that rare insight. My role at Lloyd's meant I got to look at every syndicate, what they do, how they do it, how they perform.
I really do know what good looks like at Lloyd's. And Talbot really is a very good syndicate. Yeah, look at its, yeah, look at its results. Combined ratios below 80, 10% CAGR. And still plenty of room, plenty of room through, you know, we are recognized leaders in a number of specialty classes that we really like.
Combining, you know, for us, having that presence at Lloyd's with the power and the reach of AIG behind it, not just the balance sheet, but the expertise and the capability, how can we deploy all of that around the world? It's a huge advantage for us around the world. And we use it. Yeah, it's a really critical part. And we were talking about specialty earlier. We have a really unique franchise in the U.K. We have a U.K. AIG platform. We have our Talbot at Lloyd's platform. We have our global specialty headquartered there. We've got our multinational there. Using all of those four together, it's phenomenal. The reach, the distribution, the advantage it gives us is not to be underestimated.
You still buy into it, still sees the world's risk. Yeah. It's like the only place, yeah, in the world that has access to all the risk across the world. If you have stamp capacity, if you have a risk appetite, you can find the distribution into Lloyd's.
Yeah. I mean, the talent pool of specialty underwriters and claims in that marketplace is like nowhere else. The pool of specialist brokers there. There are reasons, some of which I fully understand, some of which I don't, because I think they're tradition and emotion, actually. Specialty risk finds its way to Lloyd's. It's specialty risk and the type of client that we love, that whether we write it in specialty, whether we write it on the Talbot platform, it gives us access to risk that we would not otherwise get. In a really efficient way as well.
I think that with our scale, underwriting expertise gives us real advantages.
Absolutely. Yeah.
Thanks, John. Don, let's go back to North America and let's talk about Gladfelter and programs. It always just isn't, you know, sometimes large enough for us to get out on earnings calls, but it's a really important part of the business. We acquired Gladfelter in 2018, totally overhauled our program business. Can you provide a little bit of insight how you think about that business, how it's changed and how we should think about it?
Yeah. When you, and you touched on this again, your comments earlier, Peter, but when you think about AIG in North America, you think about the brand being large, complex, global risk. I think it does surprise people when you put up on the slide that, you know, there is a split there between what we do in large commercial versus middle market and small commercial.
With Lex, with Gladfelter, with AIG programs, we are very strong players in North America in the middle market and small commercial space. Gladfelter and AIG programs themselves are about 14% of our net portfolio overall. The average premium in each of those two businesses is $20,000. You can see where we play in that space. Gladfelter, 2018 acquisition, stable, incredibly stable, well-performing, very consistent, 75-year-old organization. It's an owned MGA model.
The basic model is that they deliver what now are very long-term, well-established programs through long-term, well-established distribution partners. They tend to be multi-line programs into what we call four, like community-based verticals. That works for us. They have been very successful over the course of many, many years. AIG programs are also programs technically, but a different story, a different journey, much more of an overhaul story in terms of that journey. There were risk class issues historically.
There were distribution issues historically in that business. What we did was took a lot of the best practices and the standards of Gladfelter and applied them to AIG programs. As we sit here today, that portfolio and AIG programs is very strong and poised for significant, you know, additional growth. It is a third-party DUA model. It is a bit of a different model, delegated underwriting authority.
We tend to work with highly resourced program administrators who have expertise in underwriting or specialty technology that we find very effective and very efficient. That journey has been one of how we actually do more programs with fewer partners. Five years ago, we had 24 programs with 18 partners. As we sit here today, we've got 30 programs with only 13 partners.
We're making that journey and it's working. It's a strong business for us going forward. With the two of those businesses, Peter, you know, what we have is rate retention stability, which is very powerful in our portfolio. We've got strong diversification, as we talked about, in terms of product channel and segment. We've got a very predictable growth engine, which is attractive for us being a part of the portfolio. We're very excited about that segment of our business overall and the profitable growth prospects of both Gladfelter and AIG programs.
It's a great story. You know, Gladfelter came in, set the standards, what we're going to do in programs. You rehaul basically the underwriting strategy to sort of mirror that. And they both elevate at the same time. Yeah. I think it's just a tremendous outcome. It's really well done.
Yeah. We're coming back to the first question. We're able to be thoughtful in terms of how we're distributing those programs in the marketplace and being very synergistic in how we're doing that. That's very disciplined. All right. We have one more question. John, I'm going to go to global personal. I think I recall the conversation is like, "Congratulations, you're promoted." And I'm going to give guidance on investor day that you're going to deliver 500 basis points or more of profitability improvement.
Yeah. I remember that. Yeah.
Yeah. Me too. Maybe just take us through, I had mentioned in my prepared remarks, it's a really good business. We like it. There's a lot of great aspects to it. Why you're confident we can actually deliver, you know, what we've outlined over a three-year journey.
Okay. Thank you again, Peter. You're welcome. Yeah. Yeah. I mean, you said it earlier, right? I mean, so firstly, I do like this business. Let's be clear. You said it earlier. We've got all the components of a really good business here. We just need it to perform better.
We are all promising, not just you and me, the whole organization is behind this. It will perform better. When we look at this, this is a new global segment. It is different portfolios at different stages of evolution and different stages of performance. We see opportunities to improve everywhere, which is great. We also see some portfolios that are performing terrifically.
We just need to turn the dial up and get even more of that. Part of our improvement is fixing the bad, but it is doing more of the good as well. If you look at the portfolios, high net worth, well talked about, we are on a multi-year improvement journey. We are starting to see the results of that come through. There is still a long way to go, and we are committed to do it. ANH. ANH is a brilliant business.
It combines every year less than 90%. We just need to do more of it, and we'll do more of it. You look across warranty, across auto and home. There's a whole mix of portfolios there. Some doing well, some doing not well. Some need remediating. Some need scale, actually, to build them up into something meaningful. We recognize that, and there's actions everywhere.
As we pull it together, you know, we've now got a global operating model for this. This shows us, yeah, they're different businesses, but there's a lot of common themes that run through those of where we can improve them, you know, and through growth, through loss ratio, through expense ratio. Growth, I just put out, yeah, ANH, it's almost like an annuity. It's so predictable. It's low volatility. It's short-tail, fixed-limit products. It's a, we do it really well.
We just need to really grow it more. That is where we are focused. Yeah. Within high net worth, focusing on E&S, which performs much better and growing that, outpacing the rest. We are already doing that, 26% growth last year on it. Loss ratios, yeah, I will go to high net worth again. Rating above trend still. Changing the business mix, not just through E&S, through other things.
Auto and home, exactly the same levers through rate, through forcing that business mix change. All of those actions recognized and underway. Expenses through our GOE. You and Keith have talked a lot about what we are doing on our expenses. That applies here as well. I think we can turbocharge that here. Acquisition costs are a big part of our expenses as well. We pay too much in some places.
We have already, you know, renegotiated the terms with PCS. We will start to see that earn through. We are doing that across the portfolios everywhere. Those common themes, that laser focus by bringing it together globally, maximizing the good stuff as well as fixing the not so good stuff. That is what gives me the, yeah. It is really helpful context. We are very focused on repositioning this business for more long-term profitability.
Yeah. Absolutely. Okay. Great discussion. Thank you. Thanks very much. Gentlemen, for, you know, giving us a little bit more detail. They will be back up when we do Q&A in the final panel, and we can dig in even deeper if we would like. In the meantime, thank you very much. Please join me in thanking John, Don and Charlie. Thank you. Well done. I would like to welcome Claude Wade to the stage. Claude is our Chief Digital Officer, Global Head of Operations and Claims.
Good morning. I'm pretty sure I have the longest title in insurance. You can just call me as things keep getting lobbed on. I am excited to walk you through how AIG is building for the future. You heard Peter, John, Don, and Charlie just talk about our growth drivers. My role is to enable and accelerate that growth through process optimization and also through technology.
It is what everybody's talking about these days, specifically generative AI. I want to caution you. This will not be a hype presentation about GenAI. We are not a technology company. What we are is an underwriting and claims company. What you're going to hear from me is our practical application of modern technology to overcome business challenges and accelerate growth.
Now, the insurance underwriting process, unfortunately, has remained largely unchanged over the last 50 years or more. AIG is not immune to those challenges. There are several significant inherent inefficiencies that exist that challenge scaled growth. The lack of data standards is the one that has to jump out at you. The insurance industry is one of the few within financial services that does not have a national regulator.
Therefore, there are no agreed-upon data standards, which leads to extreme data heterogeneity across our broker partners and results in expensive and error-prone manual data entry. This issue gets further exacerbated by inconsistent data exchange formats, PDFs, which traps data, non-standard Excel files, and even our legacy technologies are hard to get data out of. However, even as we leverage modern technology to overcome these challenges, our complex business underwriting relies on human expertise and experience.
That is the AIG underwriter. We set out to leverage technology to turbocharge our highly experienced knowledge workers, not to replace them with technology. In today's process, brokers send in submissions, which are a compendium of documents that provide underlying facts about the insured organization. The examples include the application, statements of value, property locations, to name a few of the documents that we get.
Now, we have an army of people that take those non-standard documents and go through them looking for the hundred or so data elements that our underwriters need to assess the exposure against our risk appetite, our underwriting guidelines, and then to price the coverage appropriately. The underwriter then needs to query our internal systems, looking for history with that client, claims history, and the like.
Finally, depending on the business, we may need to go externally and do a third-party search for relevant data about the business. As you can imagine, this is a highly manual and error-prone process that, on average, across our businesses, takes three to four weeks to complete a submission. The results are substandard data quality, inconsistent assessments of risk. Unfortunately, as Peter mentioned earlier, we do not consistently get to every submission we receive.
That leaves profitable business on the table. With those challenges in mind, we launched a generative AI initiative with one single objective: to unlock and accelerate growth at scale through the timely review of every submission we receive. We've taken a disciplined approach to applying GenAI to the underwriting process through five key build principles. I'm going to highlight a few of those build principles.
First, we targeted a single GenAI use case. This is something that Peter talked about earlier, focused on solving real immediate business challenges rather than experimenting with GenAI in non-core areas of the firm. Second, we co-created the solution with our business teams. This is not just an IT or a data science project. We convened an underwriter council, and then we co-located them with our digital teams so we could ensure that the solution we built aligned with real-world workflows.
Third, we adopted a human-in-the-loop philosophy. The underwriter is at the core of our business and at the core of our GenAI strategy. We set out to turbocharge the underwriter, our knowledge workers, not to replace them. Our goal is to enhance their human expertise.
Fourth, we adopted an agentic modular architecture to both easily integrate and adopt new technology solutions because this is a quickly developing technology space. We were able to capitalize on the AIG 200 investments Peter mentioned earlier in data, cloud modernization to accelerate and build our foundation. This foundation will help us continue to rapidly evolve GenAI as the regulatory and technology landscape changes over time. Lastly, our investment in these new tools is measured against predefined data metrics.
Earlier this year, we proudly launched an industry-first end-to-end underwriting assistance, executing on our idea, our concept to full deployment in about 10 months. AIG underwriter assistance is in production in financial lines and our private and not-for-profit business. With underwriter assistance, we've eliminated the underwriter's busy work. They now arrive at their desk to find each submission has been ingested, augmented, and prioritized.
All the pertinent data, unstructured qualitative data from those submissions that I talked about earlier, and the structured data from our relevant internal sources are all immediately available in a curated summary ready for them to begin underwriting. At the heart of AIG Underwriter Assistance are three key capabilities operating at scale: ingestion, augmentation, and prioritization.
For ingestion, we use large language model-powered document classification to identify and categorize the submission documents, like application, financial statements, and loss histories. We do that consistently at an accuracy level of 97% using the models. Using data extraction tools, we capture the specific data points, such as revenue, number of employees, or physical location. This leverages very context-specific language models like Anthropic's Claude 3.5 and Palantir AIP. Next, we augment the data with our own internal data, such as our policy and claims history.
Then, with third-party data and external research from trusted sources like Dun & Bradstreet and PitchBook. Finally, it is prioritization. We apply AI and machine learning to estimate the propensity to bind. What that means is it is ranking the submissions by their likelihood to convert from quote to policy. This allows us to direct the underwriter time to where it is most likely to generate profitable business and to align with our broader portfolio strategy you heard Peter describe earlier.
Together with these three capabilities, ingestion, augmentation, and prioritization, they address the biggest hurdles in the underwriting process: data heterogeneity, manual data entry, time-intensive manual reviews, and inconsistent risk assessment. AIG Underwriter Assistance is live in production, North America private and nonprofit business, and has reduced the underwriting timeline from three to four weeks per submission to less than one day.
With that extra capacity, we are now reviewing 100% of every submission that comes into our business without the need to add additional underwriting capacity. Data quality has increased from our manually input data to over 90% when we're leveraging underwriter assistance. Most importantly, our bind-to-submit ratio has increased from 15% - 20% in this business. It is fueling organic growth without the need to increase submission flow. We are now rolling out underwriter assistance to North America, all business lines, and international with a target date of December of 2026. Why should you care about this?
As Peter mentioned earlier, underwriter assistance will accelerate organic profitable growth, both within our current submission flow and as submission flows increase over time, giving the underwriters more capacity, 2x, 5x, as Peter described earlier. Now, AIG underwriter assistance goes far beyond underwriting.
The three underlying capabilities of ingestion, augmentation, and prioritization, powered by the agentic ecosystem I described earlier, are applicable to another key part of our business, and that's claims adjudication. Later this year, we'll be launching AIG Claims Assistance, which leverages the capabilities that we built in Underwriter Assistance to accelerate claims adjudication and payments, improve claim outcomes and fraud, and improve the client experience, all designed with our philosophy of human-in-the-loop, turbocharging our claims knowledge workers and not replacing them.
Now, in addition to helping enable AIG's growth agenda, these products have a profound impact on our stakeholders. For our customers, they're going to receive multiple benefits: better accuracy in their policy and data coverages, more accurate rating and pricing in their specific risks, and both improved responsiveness in both underwriting and claims.
Our distribution partners do not have to change the way they interact with AIG, and yet they will have a greatly improved experience. We can meet them where they are on their own digital journeys. They will get a response on all of our submissions quickly, thus seeing a near real-time transparent view of our risk appetite. They will receive faster responses and more consistent underwriting decisions and pricing, thus translating to a more productive relationship.
Finally, our employees. Our employees will do less of the work they do not want to do, such as data wrangling, reading through hundreds of document pages, and do more of what they want to do, such as analyzing risk, engaging with our broker partners, and simply doing the work that they were hired to do. This is a win for all of our stakeholders.
Our agentic ecosystem approach to designing AIG underwriter and claim assistance has resulted in three key strategic advantages. Let's get on the right slide here. There you go. One, we now have componentized our architecture so that we can evolve alongside any new AI technology as they come into the market. Two, we have an architecture that will rapidly scale to new lines of business, products, and geographies.
Three, we've assimilated a best-in-class solution for our unique challenges across the data orchestration and the application layers of our architecture. Peter mentioned this earlier, so I want to take a few minutes on this one. We've all heard about some of the known challenges of large language models. Even powerful models like Claude 3.5 can produce inaccurate, outdated, or fabricated information, often called hallucinations.
They can struggle to cite exactly where the source of the information provided or how the answer was derived. We have designed and implemented multiple patented technologies and a custom RAG framework to deal with these challenges. What's a RAG framework? It's an acronym. It stands for Retrieval Augmented Generation. It's designed to help large language models improve the accuracy of their responses. Let me break it down.
Retrieval, the process of finding and extracting relevant information from an external knowledge base. Augmented, the large language model, the knowledge is enriched by the newly retrieved data, thus enhancing the model's overall context. Finally, generation. The large language model then produces an answer based on its internal knowledge and the newly retrieved information. Let me bring it to life by walking through a real example.
Let's say underwriter assistance is looking for clients' total 2024 revenue, which is one of the hundred or so data fields that we need for underwriting. In our RAG framework, we would generate a prompt and a query for total revenue and then send both to our corpus of underwriting and operational knowledge. The query then returns the relevant context, in this case, the 2024 income statement, thereby limiting the vectors the large language model can actually search for the answer.
Next, we send both the prompt and the relevant context to the large language model. The large language model then generates a response, including the source document chunk from which the 2024 revenue was found, which is then reviewed, confirmed, or corrected by our underwriter, or as we call them, the human-in-the-loop.
This approach substantially minimizes hallucinations by limiting the source context for the large language model, but it also drives better efficiency in terms of response time and token cost because we're only sending the most relevant data and context to the large language model. As a core design principle, the human-in-the-loop has to validate the response and determine the next best underwriting action. I started my presentation by saying we're not a technology company.
We're an underwriting and claims company. To create underwriter assistance and claims assistance, we built an agentic ecosystem of best-in-class technology capabilities. Anthropic's Claude 3.5 is our large language model. Palantir Foundry AIP supports our overall orchestration, prompting, data layer, and governance. It all sits on AWS Bedrock and the AWS Cloud. This agentic approach gives us the flexibility to adapt to the rapidly evolving landscape with minimal effort.
As the technologies and regulations evolve, we can integrate to this new capability really easily. We're extremely excited about underwriter assistance and claims assistance and the solutions we've built to enable and accelerate growth, but we're equally excited about the ecosystem partners that we've assembled. These partnerships will ensure that AIG remains a leader in AI-powered underwriting claims and processing into the future.
With that, I'd like to invite AIG Chairman and CEO Peter Zaffino back to the stage and also invite a few special guests, Anthropic CEO Dario Amodei, Palantir CEO Dr. Alex Karp to join Peter on stage to discuss how generative AI is transforming the future of insurance, underwriting, and claims in a segment moderated by distinguished CNBC anchor Sara Eisen.
It's a pleasure. I can't believe you're willing to be seen in public with us.
Sara's going to control everything.
Sorry, I did not want to roll down the stairs, so I had to go a little slow. I am excited to be here today. I met Peter a few months ago, maybe. I went up to see him, and after speaking to him at length, I realized this is a very different company than the AIG that I remember and that we have covered for many years.
When he called me and said, "I have Alex Karp and Dario Amodei coming," I thought, "Not only is this a different company, this is a really cool company." Thank you for having me here today. Peter, I mean, how did you get all this together, and how much are you focused on GenAI and your business with these two these days?
As Claude said in his presentation, they've become very important partners. I started with Alex five years ago. We were exploring—he'll explain it better than me—but we were trying to explore how do we extract data. Here's our broader strategy.
We just regularly continue to evolve together as partners. You never know what the World Economic Forum actually was—you mentioned this earlier in my comments—like some meetings are good, some are not. My meeting with Dario was unbelievably good. We were talking about the same things. We started to learn more about what he was doing with large language models and just started to align more in the partnership. I think that both Alex and Dario see what we're doing, see the strength of the partnership, and I gave it a shot.
Yeah, there's a lot of speaking about it.
I was speaking at our investor day, and they were both there and wanted to do it.
I was going to say it was a match in Davos. It does not always happen. Alex, you--I mean, a lot of people think of Palantir, or they think of what you do for the military and for the government. And increasingly, I mean, you have been booming in the world of enterprise. How does a relationship like this start for you, and how do you think about it?
First of all, I'm honored to be here. Great partner in both cases. We started, first of all, five years ago when we started talking, the world was very different. Our products were built presupposing large language models, but without their existence. They were performative if we held them up five years ago, but now done in a very special way. They're transformative. You see it in our numbers and commercial, which is less important, but you see it.
What basically I think you've done excruciatingly well, and I think is a cornerstone for US enterprise, which is the only basically where the action is, is you've basically said, "I'm going to see if the core parts of my business or the most important part of my business, can you have simultaneous revenue growth and better economics, meaning lower cost of going to market while having radical growth?" If that's possible, that will change the way in which everyone does business, most importantly, the way I do business. That, especially given where you started before the curve, took a singular and somewhat prescient focus.
Long story short, there is a famous line from the most famous philosopher that is unknown to most people, Wittgenstein, where he said, "To follow rules—to say you are following rules, not to follow a rule." A lot of people talk about doing these kinds of things under the rubric of GenAI, and then there is really having very performative models, which is absolutely crucial. In reality, the tricky thing about this revolution is a lot of this stuff is not working. Investors are like, "Maybe it is not real." This is really working. What you are going to see is 4-5X growth in the core part of your business with, I believe, half the cost currently.
It's not just productivity, it's growth too.
What makes a business strong is, I do not have to tell anybody, is like in software, we are focused on not just growth, but quality of growth. You have these rules, and that is what they track it. The thing I would say to people in this room is this is important to look at, to learn from, because if you do not learn from this, you will see that some enterprises you are invested in do this correctly.
The delta is not just on up, it is on the quality of the revenue. Underwriting, and then I will leave it to you, is one of these use cases, which is incredibly difficult to target because you need the underlying data structure to work. There is a product we have. You need to then manage the large language model, and you need a highly performant large language model.
If you do this, you can get the 5X output with half the cost. Think about what that means where other people do not do it because they are buying things that do not work. It is really a tale of two cities. We have some other deployments that we cannot talk about in government. Same thing.
Enter the large language model. What is it that you do? How are you changing underwriting?
Yeah. Looking, I guess, from starting from kind of the technical perspective on the generative AI side, the technology is already capable of incredible things. There's this huge overhang, much more than is being applied to the economy. If we look at something like all the unstructured data that an insurance company has, the models are already very good, can already perform many of the key tasks with reasonably high reliability, and they're getting better every day, every month.
The real trick, the real differentiator is not in the technology as fast as that's improving, but finding ways to deploy it in enterprises, finding ways to revolutionize an existing business. What we found over the last few months, as we've been talking, is that AIG has really leaned into this area. They've picked one or two use cases that they really have conviction in.
Folks are adopting AI across the enterprise world, but often it's, "Oh, we're going to try this. We're going to try this. We have a bunch of pilots." They really had conviction on the claims cases and the underwriting cases and have moved quickly to get a lot of this unstructured data in and really bet with conviction. They've been using Claude for a while, all the way back to Claude 2.1, which I think was late 2023, early 2024, was it?
Now things are really ramped up with the new, stronger models. The technology can be great, but if there isn't that conviction, if there isn't that will to move quickly and that focus, it doesn't happen. This is one place where we really have it, and that's where I feel good about the partnership.
How does it change the business of underwriting, Peter?
For us, it's going to be getting us access to data more real-time that we just don't have the ability to harness today. An underwriter, by their very nature, wants to look at as much data as possible to make a good underwriting decision.
Where we've started is in the data ingestion, and it comes in from our distribution in so many forms. With what we're doing with Foundry, you can take a PDF and convert it to text. I mean, you can take some things that we just weren't able to get to the underwriter for insight. With what we're doing with the large language models, you get it in a fraction of the time. There was not necessarily early days as much belief.
We said to the underwriters, "I'll give you an infinite amount of time if you'd like." We did. Within two to three weeks, they said, "Okay, I have what I need." They got probably 75% of the data that we had outlined. With what we were doing with Palantir and Anthropic, we got 92% of the data in three hours. All of a sudden, it was a shift and a dynamic that is cultural in that we're going to not waste time in underwriting and inputting data. We'll have more data, more insight, and better decision-making. It is going to be profound for us over time.
Are your competitors doing this too?
I got that question from someone in the audience. I do not think so, but I do not think it much matters. I mean, I think this is the way it is going. We talk about both Alex and Dario have conviction in this three to five-year period, almost more than the 12 months. I do not want to speak for them, but we know this is the direction and the way business is going to be conducted. We are just fully committed, but I do not believe any of our competitors are doing end-to-end like we are.
Alex, what have you learned about deploying your software and your tools, your AI tools for insurance and financial services more broadly?
You have to just reframe what you just said. If you start with a simple statement, can AI outperform a human in what they do in the following way? Can AI make one human, five humans? That's basically if you want to. Here, the human is doing something very technical.
We talk about underwriting, whatever the use case is, you have an expert who's been trained in understanding not just data, but proxy data. It's very hard to teach people proxy. You guys are all watching this, and you have the people who are very good at what you do. You're not just looking at what I say. You're looking at inferences for what it means based on years of experience. That's how you know if what I'm saying is actually going to change your model of their business.
Five people say the same thing. One is convincing, one's not. That is how you make all your money. You can tell the difference. You can hear what people—so that is de facto what a technical human is doing. Can the large language model combined with what we call our ontology make one human 10X more valuable, meaning they have 5X more output at half the time?
That is actually a crazy hard thing to do because you are going to have to figure out where you deploy the large language model across the decision tree. It is not one decision, just like I am making many statements in your bracketing. That makes sense. That turns the dial. That changes something. That is different than what I heard. That is what I have heard before, but because of that, it is commodified. I do not believe it.
Those are assessments you're making. Underwriters are doing exactly the same thing. It is a huge milestone if you can actually come in and change the unit economics dramatically. Dramatically is important because there's a huge institutional cost to this. The question you're going to ask yourself is, okay, or many of you are going to say, "I assume they're, I believe them.
He's a little crazy, whatever." You're going to, de facto, the next question is, is this a commodified way of doing things? What I would tell you is it would be if you actually could do it. It is going to be very hard to do it because first you need a CEO who's like, "I'm going to go to core principles. What is the core principle of my business?" Many CEOs do not have the agency to do this.
Whether they believe they have the agency to do it or not, they can't do it either constitutionally, intellectually, or because they're owned by someone who doesn't give a fuck. You have somebody who's willing and able to do it, and you have somebody who's willing to basically talent spot and say, "Okay, this could work. Let me see it every step if it's working."
People on the other side of this are going to be slow to adopt because they're going to have to get to the other side of the transom and see the results before they're willing to put up with the pain of doing the culturally internal changes. That is a huge bear to injury simply because what we're going to be doing together in six months is going to be dramatically better than what we're doing now.
It's like you get better and the models are already better because we know the use case. Our way of managing the models is dramatically better, and it's not easy to replicate. Not easy, I don't want to say impossible, but it's very, very, very hard. You don't believe me on this. There are very basic things like why are we both American companies? That doesn't make sense. It's because there's a lot of cultural, tribal, and basic ways of doing these things very hard to replicate.
Do you agree, Dario?
Yeah. No, no. I mean, I definitely agree. I think, as I said before, I think getting this stuff actually into the enterprise, connecting, it's almost like there's this impedance mismatch you have to solve and that we're solving some places faster than others, which is you have this incredible technology.
You can do a demo. You can see it's capable of incredible things, but actually plugging it in, actually making the two things connect, I think that has enormous mutual value for both sides. You're not just selling electricity. You're not just selling water. You're kind of creating mutual value. I mean, one thing I think I saw is if I look kind of across the enterprise world, there's sort of like I would call it the periphery and the core.
I think enterprises are having a lot of success deploying in the periphery. That's kind of common across all companies. All companies need customer service. They need internal productivity for their developers. That's the same whether you're in insurance or some other area. A lot of success there and a lot of aspiration to make progress in the core of what they do.
There would be underwriting for insurance companies. It would be drug development and clinical trial prediction for pharmaceutical companies. It'd be the core of what legal companies do. For a company that does trading, it would be kind of trying to figure out the key algorithms. In many cases, more aspiration.
By the way, this is a very important point, and I do not think it should. I'll translate this in my more vulgar language. Lots of companies doing stuff that does not matter. Does not matter at all. You commoditize largely because a lot of these things are, I do not want to call fraudulent, but it's like it does not matter for your core business. The impedance mismatch here is really big, and therefore people do not have the courage to go into the core part of the business and say, "I'm going to change the unit economics of the part of my business that drives the business." Because that's really hard.
Why? Because of cultural?
No, because it's technically very hard. I mean, it's a long. The models are capable of it, but just every little detail of it. There's a slight. Particularly in a regulated industry, there are lots of challenges. You guys are confronting those challenges head-on, which is why we have this conviction that we're going to make it work together.
I think this is like a heavy-duty compliment for you, Peter.
I don't need to say this again.
That you're getting, these gentlemen. Was it a tall task?
I don't think it's just regulated, by the way. I think even if you take that out of the, it's like there's, first of all, you have the regulation. Then you have the core security models and the way the data is modeled on the inside. You have where do you deploy the model and under what condition. You have how do you train the model and to what use case. Each one of those would require a really world-class player.
The minute you don't have one of those things, the whole thing collapses. People have figured this out, which is why your phones are ringing off the hook for like, "Hey, we can help your call center." It's like, "Yeah, we can make your call center 20% better." That's great. That does not change your business.
How did you do it? How big of a lift was it? Either culturally, technically, all these challenges that other CEOs have.
It's been a significant lift over time, but what we did several years ago, not knowing we'd be here today and talking about large language models or how we ingest data differently, is just getting foundational end-to-end process, which the company did not have, built and get single definitions of data. For us, we were committed to changing our company, and that took a lot of will.
Once we started to get there and introduce new ways of doing business, there's always reluctance in big companies. I mean, I say you have to bring an industrial weed whacker to work every day. I mean, if you don't beat back the weeds, it's going to take the house over. We kept pushing. We didn't roll it out to the whole company because we weren't looking for buy-in. What we were looking for is learnings.
I think working together and seeing what was possible and how that worked for us in the business that we were in became really aspirational. We just made that commitment, made the conviction, and we have been enormously pleased with the progress that we have made. I think the future is even brighter.
Dario, question on Anthropic and Claude. Why Claude? Why is that uniquely positioned for doing these underwriting tasks? What's the edge with using you over OpenAI or Gemini or any of these?
Yeah, a few things. One is I think we have the best models overall. There's some general factor of model ability and intelligence, and Claude models are arguably the best on that axis. I mean, people release models all the time, but we've been on a very positive trajectory, and we have a lot of conviction that we're going to have the best models in the future. I think separate from that, we've focused a lot on things like trust, safety, reliability, and responsibility.
That's particularly important in these regulated industries where you have to make sure that your customers trust you. You have to make sure that you're doing the right thing by regulators and by society. There are several levels at which we've focused on this.
Our models are trained with this method called constitutional AI, where they're trained with a set of principles that they operate according to. This makes their behavior more predictable, more reliable. We've always highly valued privacy and security. Some of that is with our cloud partners like AWS, but it's always something that we valued ourselves.
A new emerging area we've actually been working on for several years is the ability to see inside the models and make their decision process transparent. There's a field of research called mechanistic interpretability that's just starting to get commercialized that allows us to look inside and trace the thinking of the model and understand why it makes the decisions that it makes.
Of course, in things like insurance, like underwriting, like claims, it is very important to society that we be able to understand what is going on inside these models and that we understand why they make the decisions they make and that those decisions are fair. There is no other company that has made anywhere near as much progress in this area or that focuses on it or holds it as a value as much. We are very excited to use these methods that we have developed to bring them out into the world and to apply them to many of these important applications.
Do your competitors not value trust and safety?
I mean, this is one of these things where everyone will, of course, say that they've.
You've been there. You've been in the spinhead.
If it sounds like a good thing, everyone's going to say, "Oh yeah, we have that. We do that too." I think just the research and the deployments and how fast we're growing in the enterprise and developer space where trust and reliability are at a premium, I think those things speak for themselves in terms of who is really serious about these things.
Yeah. I mean, Alex, how do you do that? How do you build a culture of trust and safety while also just moving forward and fast with innovation?
Let me just reframe the question and the answer. Let's just assume that you don't trust anything we're saying. Maybe a different way of saying it is by raising the complexity of the problem. In the anti-terror context, it would be civil liberties and killing terrorists. Y ou raise the technical challenge, and then the output of the software, the quality in aggregate becomes better, even independent of the claim you're making.
A different way of maybe answering the question is, maybe if you espouse safety, the underlying model becomes more performant because the challenge you're trying to solve is harder. The engineers have to be better. Your culture has to be more focused internally as opposed to just delivering what the client says they want tomorrow.
I think that is actually the real answer and the secret to what we've done is you raise the complexity of the problem to the point where it's nearly impossible to solve. By solving it, you solve lots of other problems. The model is there's always a question of what do you mean by intelligence. All these things are proxy indicators. Is IQ really an indicator for success? It's kind of a proxy indicator, but it's not a.
How do you have software or models that are more performant than even the test you would give, so the ELO score? It's by raising the standard. In our case, it's like we built this thing called ontology, which allows you to microcontrol inputs, whether those inputs are humans or they're from a large language model in the institutional context that is mandated.
It is really because if you take the next logical step from what you would need to do something, you end up with a series of products. I would say, again, I do not want to be ridiculously only pro-American here, but this is like a very American way of solving things, which is like, let's bring fairness and integrity into the core problem, which is probably how you ended up assembling us.
The aggregate output, the sum is much greater than the parts. If you try it the other way, you get a very dysfunctional thing that is not strong enough to stand up on its own legs and certainly will not work in enterprise. I mean, from the outside coming into an enterprise, going into enterprise is always very difficult.
Your value proposition has to be dramatically better than what an enterprise can do itself because it's painful for the enterprise to bring you in, even if you have the best sponsor in the world. That's how you do it. Actually, the way you build software is exactly what you did.
You took a micro thing and said, "Let me see if this works." You use that thing to convince your people because if you had just said, "Hey, do this," they would be like, "Well, why?" It spreads. You put more gasoline on it. It's like it's cutting the weeds, which you're right about, and you did a great job of that. It's also planting seeds and making sure they can grow.
On that note, Peter, what are some of the problems of the future, problems that you deal with in the business that you can have these guys try to solve? Climate change? I don't know. Fires in California like?
Yeah, there's a lot of that. I think the next evolution is we have really focused on the front end of the business, which is data ingestion, getting underwriting criteria, training the large language models. I think how to be more nimble and allocate capital and optimize a global portfolio more real time, which you'll need more computing power, but you'll also need ways in which you can develop large language models to know how you're going to take a portfolio and apply it to individual customers.
I think that's the things we're talking about with Alex and Dario as a next step. I bookend myself with really smart people. What I do know through this is that, and I talk to Claude about this all the time, is what we knew 12 months ago, I know so much more today.
I assume in the next 12 months, I'm going to know a lot more than I know today. Having that sort of mindset that we're just going to continue to build foundational work and try to adapt to the advancement these guys make has been fantastic for us in terms of advancing our business.
How do you see the future of the partnership and just in general, where are we going with all of this?
Yeah. I mean, I am just excited for the rate of progress of the models. I have been following this for 10 years since I was an individual researcher working on the models. It is incredible the regularity with which the autonomy, the ability, the accuracy of the models just gets better and better every year. We have no guarantee. The music, in theory, could stop at any time, but we now have a very long track record of the models fundamentally getting better at what they are doing.
As they get better at what they are doing, you might think, "Oh, they are smarter, the complexity goes up." Actually, the more autonomous they are, the more the deployment friction goes down rather than up. If a model is very weak, you have to put all this superstructure around it to make up for its failures.
As the models get stronger, our ability to deploy them for things that are more and more core to the product, more and more core to the underlying product and the underlying financials of the business in question, that is going to be very much more possible to do.
Of course, we will mutually learn a lot more about even if we were to hold constant, even if we were to freeze in place the quality of the current models over years, we will learn more about how to deploy them. We will deploy them much more broadly. I am sure we will make some mistakes. We will learn from those mistakes. We will iterate rapidly, and we will produce great results. You put this all together, and there is a lot to be expected.
It is not hitting a wall, contrary to some.
Very, I would say very unlikely. I don't speak in certainties, but we have a very long empirical track record of evidence against that.
Is it going to need less compute, which is kind of a hot question?
Yeah. I mean, I think the dynamic we've seen is that to make the cost of making a model of a given intelligence quality and capabilities is going down and down and down. The amount of money that we're putting in to make models is going up and up because the models are so good at what they do that they're so valuable that it just makes sense to put even more effort into training them.
Final word, Alex, how do you see the future? What are you thinking about capabilities that you wish you could do now with the technology that's not there yet, but that's where we're going?
There is a lot of debate around this, and I would say I go back and forth. I do think there is a separate question of not just how good they are, but how reliable they are at the point of use. You could have the models getting much better, but they only work 80% of the time. There are some statistical issues there. What I like currently is, first of all, it is a revolution born and bred in America, the bastion of all good in this world.
We're all about America.
I do not know if we are all about that, but everyone should be.
You are.
I'm trying to spread the wealth of wisdom. And two, you're going to see in the next two years, let's just figure not five years, next two years, a lot of theory is going to move out. We've had this by the Dario never did this, but there are some people in the Valley who've sold more than they could do. And the theory of AI is going out of the market, and the reality of what works is coming in.
That's very good for Palantir; it's as good for everyone on the stage. I'm looking forward to it. I think it's crazy important on the battlefield, a subject not for this conference. And the concern I tend to have about AI is much more societal.
It's going to make, I do think it will lift boats, everyone's boat, but it's going to lift the boats of people who do it well disproportionately, and that's going to create societal and investor issues. Getting these things right is going to be crucial to everyone's life.
What's the best way to see who's doing it well? Besides using Palantir and Scrap?
No, no. No. Do not know exactly. Start with nothing. Do not assume anyone's and then go like, "Okay, I would assert this is one of the strongest implementations of AI you can find." Great. Invite yourself in, take a look at it, and then invite yourself to some other people who've spent more time, more money, better at propaganda. Compare the results. Do not make the mistake people make.
They then say, "Okay, well, give credibility to the people who are actually doing it." The biggest mistake people make in software, and I would say also in the implementation of large language models, is it's way too theoretical. Do not listen to what people are saying. Go watch the results, and then listen to the people who deliver results, and push the people who are very good at propaganda and very good at persuasion out of your business. That is how you will do well. This is a very good example of something that is very strong. Take time, go look at it, and learn from it.
Peter, I think we'll be happy, right?
I'm not happy.
Thank you. Thank you very much for inviting me. Thank you.
Thank you so much. Yeah, we got a picture?
Oh, a photo, a photo. Peter wants a photo.
We just do a photo here.
Photo? Yep. Oh, we're going to get the arms around.
You're very brave, Alex. That's good.
All right. People roll me out in public.
Sorry, thanks. I'll see you in a little while. Thank you.
Thank you.
Dario, I'll see you in a little while.
Yeah.
I got to finish a Q&A. Yeah.
Yeah, we're doing that.
Okay, come on up. No pressure to top that. Yeah, I mean.
I'll follow that. All right, following that, we'll be tough. Act, but hopefully, we'll keep it going strong. There are microphones coming around. Again, I'm just going to ask everybody to limit themselves to one question and one follow-up in the interest of getting to as many people as possible. We will kick it off with Meyer Shields from KBW.
All right, I'm going to stand up because you told us to. Meyer Shields, KBW. I'm going to start with a question on the last panel, and that is it's going to be, well, I'll let you be political in responding. Historically, the insurance brokers have generated most all of the returns in the P&C space over time. With these tools, how does AIG ensure that it's retaining as much of the value that you're creating through better underwriting and claims handling?
Thanks, Meyer. I mean, the idea is not to change the business model. We know that brokers, agents, very strong partners of us. I do not think that what we are doing can be replicated by them in the short run or even they have to get the capital. How they submit data, they may change, they may not.
I mean, but I think what we are trying to do is be agnostic in terms of how that comes in from our partners, but changing our business model to be better service, better capabilities, better insight. Over time, you heard from Alex and Dario that the models are going to get better. I think it gives us the ability to underwrite more effectively over time. What that looks like in terms of specifics, we will see.
I mean, but I think today is a very frustrating industry in that things have not changed. Whether it was algorithms, digitizing workflow, you still can't change the way the data comes in. I think this is a, that's why we're doing what we're doing. I think this is going to be a dramatic change in terms of how business is conducted in the future.
Okay, thanks. Completely unrelated follow-up, but I wanted to ask Charlie. A lot of the slides so far have been forward-looking in terms of ROE improvement, expense ratio improvement. We did not get a lot in terms of optimizing reinsurance purchasing now that AIG is fundamentally more credible as an underwriter. I wanted you to take us through what we should expect there.
Sure. Look, I think it's really important to go back to what Peter was focused on about that philosophy because that's been the real heart of everything we've done in terms of reinsurance. I talked a bit about the cap structure earlier. What does that do? It's protecting capital. It's managing volatility.
We have an earnings lens on it as well. All of that is contributing to improving and increasing the quality of the underwriting earnings. Our view, it's a real short-term view to turn around and say, "I'm going to flex my reinsurance structure to try and look to short-term, whether it's growth or short-term profitability." Frankly, where I think you've seen people do that in the casualty space, it really shows if you think about the casualty performance.
Again, one of Peter's slides, a very consistent net retention. Casualty goes wrong on a calendar year, not on an accident year. When people flex their reinsurance structure, that tends to cause and lead to problems. I think we're pretty happy with where we are.
Next question, Josh Shanker from Bank of America, please.
Yeah, thank you, everyone, for taking my question. I guess Charles or maybe Pete. I am not supposed to engage with people who are smarter than me, who have nothing to lose, and I guess are better negotiators. Why should I be doing any reinsurance with AIG in terms of if AIG is all those things now, how do reinsurers act as partners with AIG instead of being opportunities for AIG to have a long-term relationship?
I'll start, and Charlie, you can add. I would say our reinsurers have been one of the most important partners for us in this sort of evolution of changing AIG's reputation in the market, getting to underwriting profitability. We treat them with enormous respect, with enormous transparency, and they trade across the portfolio. While we have very strong reinsurance structures, it is a multi-year approach.
Our reinsurers back AIG. It's not a transactional trade of there's property reinsurers, there's casualty. We get the world's best reinsurers to look at our portfolio. After we get things done on treaties, Charlie's right back in with our partners in terms of thinking about the strategy for the future. They have a great balance, and they know we don't trade on a transaction. It's not, "Could we push price more?" Absolutely. "Could we do things more short-term?" Absolutely.
I think this evolution process has benefited us because of the partnership approach we've taken, and they should trade with us because they will make money over time. This is not an income statement trade. It's a balance sheet trade, and that if they stay with us over the long period of time, they're going to get their returns and get their target ROEs by being part of us. Charlie?
Yeah, I think you agree with obviously all of that. I think if you go back to when we were talking earlier a bit about 2017, I think had the organization sought to buy more thorough, more thoughtful reinsurance, as I said, I think I mentioned it was very opportunistic.
That is a short-term strategy in terms of with reinsurers as well. We spend an enormous amount of time, Peter and I, really rebuilding the trust in AIG because reinsurance is not a substitute for poor underwriting. That is the key. The relationships we built, they started at the very most senior levels within our organization and with the organizations that we trade with, the largest reinsurers in the world. I think there is that mutual trust and that understanding of what they offer and how it matches to what we want and that it is there for the long term. It is not something we worry about at all, actually.
The ROE guidance of 10-13%. We knew the beginning of that coming into 2025, the 10% ROE guidance. Some people were skeptical that AIG might be able to deliver on it. Within two weeks, you had a $500,000,000 non-monthly cap loss with some additional reinstatement payments that are going to go on top of that. You are still confident about the 10% ROE for 2025. Has something changed since the year began, or were you going to greatly exceed 2025's goal when you started the year, even though we were surprised by the LA wildfire losses?
Yeah, nothing's changed. As Charlie said during the panel, the $500 million, it was modeled, it was modeled incorrectly, which is most caps, but that one in particular cut off the tail, it was much bigger. It's January. We have, back to the slide I put up on our percentage of our loss ratio on cap, has been very steady at around 5% over the last four years. It's the way we structure our aggregate.
I mean, who knows what distribution of cap's going to be for the rest of the year. I wouldn't look at the $500 million as additive. I would look at it as part of our expectation of frequency within cap for the full year. Again, not knowing the distribution, knowing it's March, we haven't hit wind season, things happen.
That $500 million is a contributor towards what we're comfortable with as retaining net with the currents and aggregate. Could it be a little higher? Maybe. Could it be a little lower? Yes. That is why we have a lot of confidence in reconfirming the guidance.
Thank you.
Next question from Paul Newsom from Piper Sandler.
Thank you. I was hoping you could expand upon the role of private capital with the SPV. Just how far can you end up expanding that across your business lines? Is that something that we should expect quickly and broadly? A little bit more color, I think would be interesting.
Charlie, let me start and maybe you can provide a little bit more color. The idea of the special purpose vehicle was an excerpt from the portfolio optimization work that we've done and felt that if we can create something with an enterprise model that shows that the investor is going to get a very strong return from diversification as well as what our expected losses are, that it would be a good match.
We've been working on this for two years. For us, looking at how we place reinsurance, maybe it's like 15+% of our total reinsurance outwards, that it's a really good way to secure capacity that's more consistent that will follow the leads and be there over a long-term period. The idea always has been that that becomes a base to build on other risk pools over time.
What that looks like, what it is, we will explore it. There are definitely opportunities in the marketplace. We figured that our own reinsurance at roughly $750 million of placement, sizable, it could be a base that we can then diversify as we see different risk pools to be able to raise additional capital where we would get fee income, but also see investors be rewarded for scale and diversification. Charlie, you want to add to that?
Yeah, I think we're, as I hopefully came across in our panel discussion, we're really proud of what we achieved, but we're really excited about what the future opportunity is. We have this incredible platform in terms of generating risk. I think we also have the capability and the skill set to manage that risk into portfolios that really are about finding the optimal capital.
I think what distinguishes ourselves maybe to where you see on the distribution trying to do it is we also have our own balance sheet. We can take the risk ourselves and shape the risk ourselves. You can't do that as a distribution company solely. We think we're as well placed as any company in the world to really capitalize upon this, and we're really excited about it.
Next question from Daniel Lee from Morgan Stanley.
Hi, Daniel Lee from Morgan Stanley. Thank you for taking my question. My first question is just in regards to the last panel. It does seem like it's a very cool collaboration for sure with Palantir and Anthropic. It does seem it's definitely a mix of build versus buy when it comes to tech improvements. How should we think about the mix of build versus buy going forward for AIG?
Claude, do you want to take that?
Sure. As I said in my remarks, we're not a technology company. We're an underwriting and claims company. We don't build technology. I mean, everything that you've seen in terms of our end-to-end ecosystem, we describe it as this modular agentic ecosystem where we're plugging in fit-for-purpose solutions that work for our unique data challenges.
That's where Anthropic, Palantir, AWS, Titan all come into play. The only thing we've built is the overall wrapper and the workflow that connects them. Outside of that, everything is a modular approach this way. We can always have best of breed, and we can grow as those organizations continue to improve the capabilities of their models.
I think probably one thing I would add is that we have, though, in order to be prepared for that, have spent a significant amount of money on our foundational capabilities, our data insight. That has been a journey between AIG 200 and then what we have advanced in terms of building out the business. I do not think we would be able to do what we are doing, as Claude outlined, with Anthropic, with AWS, with Palantir, if we had not made those significant investments to build the foundation.
It all starts with the data.
Cool. Thank you. My second question, I guess, would be for Keith in regards to cloud. You guys mentioned you guys are now at 80% on the cloud in terms of workload. In general, we know cloud expense can be a variable expense. In terms of, as you guys expand your AI capabilities and other tech workload onto the cloud, how should we think about how you guys would manage expenses in this area going forward?
Can I take that, Keith? We know there is a variable cost on the cloud. Having partners that are embedded with us with Bedrock, Anthropic, within AWS, we know the per cost for, I mean, we are able to monitor that and not let expenses sort of get away with this hypothetical experiment, but it is thought through. There will be some incremental expense as we start to adopt this more broadly across the organization. Having said that, there is a commensurate growth element to this that we should be able to absorb it.
This is not introducing this today of saying, "This is on the come, doesn't sound great, and by the way, there's going to be more expense." We will be able to manage the expense on an incremental basis in terms of what we are driving over the next couple of years and provide guidance if there is something that is going to be accelerated beyond that. We have a very good, deep understanding of how we use all of the technology and applications on AWS.
I would just add, it's why we spent so much time walking you through the individual steps and components of our RAG framework. Because we're only sending the models what they need in order to process, we've actually lowered our cost for each transaction. We expect to see that continue to be lowered as the models get better and as we get better in terms of providing the context of the model.
Thank you.
Take our next question from Dave Motemaden from Evercore ISI.
Thanks. Good morning. Just on the last panel, I thought it was really interesting just on the new technology driving growth, particularly within the commercial lines. How much do you think that that can turbocharge growth versus the 7% net written premium growth that you guys had last year? How should I think about maybe just a follow-up for Don on the ability to get more share with your trading partners within that?
Don, why don't you start with that? I'll take the first.
Yep. As I said, in North America, I'll just talk about the retail business. I mentioned the data point that top 10 or 83% of our premium volume, share of wallet, which I talked about data availability, share of wallet with them ranges from 2% to 7%. It's not like you're talking about going from 30% - 40%. It's like going from 2% - 4%. It's like, can we double our business with these top brokers that we do business with?
The question lies in to what level of connectivity and alignment have we had to date? I think there's, I'm talking about my background as a broker, I think there's a bit of a fallacy that there's been a high level of connectivity, been a highly strategic partnership. I don't think that's the case.
I'd say on a scale of 1 to 10, they've been about a 3. I think if we can get to a 5 and a 7 and an 8 and a 9, there are tremendous organic growth opportunities with our top broker relationships for years to come. I'm very confident about our ability to grow that space.
It's hard to give specific guidance on what we think the growth trajectory could be with adopting what we've outlined in quite a bit of detail here today. I think the Lexington example is a good one. We don't have to go back through that. If you think about other scalable businesses where cycle time matters, you can point yourself to Japan.
I mean, we haven't fully started there, but the cycle time of getting quotes in a fraction of the time for personal insurance, small commercial will have a dramatic impact on our ability to grow that business. Yes, you may be taking market share, but more importantly, you're probably getting more growth from additional product because you're able to actually quote that in more real time.
I'm talking about not like the example we gave before in Lexington, which could be days or the one that we just did the pilot on was weeks, and then you get it in hours. I'm talking about getting it from an hour to like 12 minutes. I mean, if you can do that and you actually have the products because it's more complicated in Japan because every product is somewhat homogeneous and then there's like 50 riders that are heterogeneous and you have to build that out.
We do think that there's real opportunities in our scalable businesses where cycle time matters to have multiple growth that we do today. What that looks like, what's the actual percentages, I think that's what will evolve and take you through over time as we start to adopt this more broadly.
Great. Thanks. My follow-up is back on the M&A framework. Thanks for that. Just on the financial targets, the EPS and ROE accretion, is there a time frame that you think about to achieving accretion? Is it immediate? Is it one to three years? Should I think about that as impacting the $5 billion-$6 billion of share repurchases that you've outlined in 2025?
I think that was three questions. Let me see.
I held back a little bit too, so.
Which one do you want me to answer? No, it does not impact what we've given you guidance on for share repurchases. Look, our base case is that it's not going to be dilutive. Would it be if we found the right acquisition in a one to three year period? Maybe. I mean, I would think about we have capital to absorb risk. There are different ways in which to structure that.
Driving more earnings is going to be accretive to ROE. We are really focused on not coming back to this group and saying, "Oh, we just did a deal, and by the way, we're talking our guidance away and give us more time." I mean, I would not want to do that. If there's something so compelling that it changes the organization forever, you'd want us to do that.
Because if you look back at AIG historically, and I'm not judging it as just math, that after it went public again, the money it spent on three buckets, which is reserve strengthening, raising debt, and then share repurchases, was between $90 billion and $100 billion during that period of time to keep the stock flat.
I mean, had they found something compelling, and again, there was a point in time, I'm not judging it, I wasn't here, but you drive earnings, you drive more market cap, you drive more option value, you're able to today adopt LLM on a bigger platform, we want to do that. We want to be sitting here for whatever period of time is showing you that we have real potential for meaningful growth. This company's built and has capabilities to be bigger. Doesn't have to be. It has great size today, but we can be bigger. Also, we've shown through the operations that we have the muscle to transform. I think that we would be a good acquirer of businesses if it meets those thresholds.
Great. Thank you.
Next question from Elyse Greenspan from Wells Fargo.
Hi, thanks, Elyse Greenspan, Wells Fargo. I wanted to come back to the reserve discussion earlier. Keith, you were talking about just the conservative nature of reserves. Within the guidance, the 20+% EPS CAGR as well as the ROE guide, are you guys assuming anything for reserve development? I'm thinking maybe just the ADC and then anything else on top of that could just be upside to guidance.
Yes. I love your questions that I get one-word answers, Elyse.
I made that one easy. The second one takes the expense ratio comments and also just the last discussions on AI. Obviously, there's a cost there. You guys have obviously been investing in AI, and it sounds like something that will continue. How does that cost get factored into the desire to show material improvement in your expense ratio?
Claude, maybe just go a click down in terms of how the variable cost works. We have a lot of initiatives in place. We talked about AIG Next earning further into 2025, getting the businesses sort of set up to be leaner, other operations leaner. We have a path there, Elyse.
One of the things that I think we do not necessarily get full credit for is that even when we were giving guidance on some of the other operational programs that we were doing, we were investing all along. I mean, we were investing in people. I mean, when we started with the data, we had nobody. I mean, everything was additive in terms of how we were actually going to position the company for the future operations. That is not this group's problem.
That's our problem is to be able to invest, reposition the business, and still meet guidance in terms of what we're going to do on the expense side. We need to be below 30%, and we're going to be all over getting that done while reinvesting for the future.
I would just add to that, Peter, this effort up until now, it's been an end-to-end transformation. I talked about optimizing the process. It's people, it's process, and technology as an accelerator and enabler. We've been able to offset the cost of implementing what you saw today through optimizing the process. That's all captured today in our numbers. We'll continue to do that.
We'll get leaner, we'll get smarter, we get faster, and we figure out how to take the waste out of the system while we're implementing the new solutions that we're building. Again, we're doing it in a cost-efficient way by that RAG framework and making sure we're only sending what's necessary to the models, which is what's driving the token cost. That's really what the cost drivers are for a solution like this.
Brian Meredith from UBS.
Yeah, going back to the AI discussion, you're doing a lot of stuff. It looks great. As an outsider, what are the things we should be looking for to kind of grade you on whether this is successful or not? I mean, the 13% ROE target, that's terrific, but I look at other companies that you compete against. They have much higher ROEs. Do you anticipate this will take you to an ROE that'll be best in class or best in class combined ratios growth? How should we measure this as outsiders? It's difficult to.
Quentin, you never give Brian the last question. How many times have I not? Brian always tries to get a double or triple guidance. Other companies, here's where I'll start the other company comparison. Nobody's had a start where we started. If you look at ROEs of different companies, that doesn't mean we're not going to have aspirations to get there. They're coming back to the average. We're moving up.
I mean, certainly with the scalability, again, there's no time frame for this, Brian. I said this is from 2025 to 2030, but certainly being able to grow in a way that we have not seen in the past will give us opportunities for ROE to be more accretive for sure.
I mean, the 10-13 does not contemplate large language models or what we've outlined today of hitting that out of the park. That would be in addition to. We're very focused on making sure that we're driving those outcomes and investing for the future. You should measure us based on the progress that we're making, not too tight of time frames, but how are we rolling it out to businesses?
How are we actually effectively changing the way in which we can ingest data, submission activity? How do we grow? How are we driving this across the enterprise? If you want to do it easier, just compare us to our competitors, and I think you'll be happy with the progress that we're making because I don't think anyone else is doing the end-to-end.
One more just quick one here. Obviously, it's a change right now. AIG now is more about growth and not fixing. Any changes to incentive compensation that we should be or you've been thinking about making changes at all, looking forward versus what it's been over the last couple of years?
Not particularly. I think one of the changes, like when you could imagine when you're first bringing in people who were going to be critical to the turnaround, that getting people convinced on the calendar year compensation wasn't easy. Those are some of the changes that we've done with our board of directors to really drive all of the incentives to be aligned with every aspect of financial performance and the company.
Now it's ROE. It's making sure, as a matter of fact, we took out because we had a two-year commitment on getting other operations down to this optimal $350 million for parent. We got there sort of 12 months ahead of where we thought we would, and we took it out because there's a new metric.
Now we're driving things that are aligned with all of our shareholders and making sure that we're driving EPS, ROE, calendar year, and accident year combined ratios. Everybody has the same incentive, and it's not different for underwriters than it is for functions. It's not different for corporates. It's for the business. I mean, it's all aligned, and people get paid if we drive value aligned with our investors.
Thank you all for the questions. I think with that, I'll turn it back over to Chairman and CEO Peter Zaffino.
Thanks. Thanks for it's been a great day, and I really appreciate all of your attention to our story. I'm going to say one sort of quote, which many have said before, that one's grasp should always exceed its reach, meaning your goals should always be greater than what you're actually capable of achieving. For AIG, our reach is our grasp. I mean, we've shown that we've had enormous aspirations.
What we've accomplished, we tried to outline it this morning, has been nothing short of extraordinary. I've been told it ain't bragging if it's true. What things you've been overwhelmed with information and a bunch of data, what would I take away from today? The past is in the past. We've taken no shortcuts. We've worked at massive pace.
We've run to problems and have worked through all the real critical issues to make sure we position ourselves for the future. I would say the best predictor of future performance is past performance. Today, we've demonstrated that we always deliver, and we want you to really bet on this team. The next, the third one would be the company has enormous strategic and financial flexibility.
We're going to be very disciplined with every decision as we always have been, but we feel like the industry is facing headwinds, and we've never had more momentum as we enter 2025. The last piece I want to leave you with, we went through it a lot today, but AI is real, and it's not a cost-cutting exercise. It's an end-to-end exercise. You need scale, expertise, and a strategy that will propel you into the future. We have that. We've accomplished a significant amount, and we think the best days are ahead. Again, we really appreciate your time today. Hopefully, it was informative, and look forward to spending time with you at lunch. Thank you very much.