Good morning, and welcome to Lemonade's first ever Investor Day. My name is Yael Wissner-Levy, and I'm the VP Communications here at Lemonade. I am thrilled to be with all of you here this morning in our New York SoHo offices and with all of you joining remotely online from around the world. I'm gonna quickly read through some necessary disclosures, and then we'll hear from our Lemonade leadership, answer some of your questions, and get on with this day. The recording of today's investor presentation that we will discuss today is also available on our IR website, investor.lemonade.com. I'd like to remind you that management's remarks on this call may contain forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995.
Actual results may differ materially from those indicated by these forward-looking statements as a result of various important factors, including those discussed in the Risk Factors section of our Form 10-K filed with the SEC on March 1, 2022, and other filings with the SEC. Any forward-looking statements made on this call and today in these presentations represent our views only as of today, and we undertake no obligation to update them. We will be referring to certain non-GAAP financial measures during today's presentation, such as adjusted EBITDA and adjusted gross profit, which we believe may be important to investors to assess our operating performance. Reconciliations of these non-GAAP financial measures to the most directly comparable GAAP financial measures are included in our investor presentation.
Our presentations also include information about our key operating metrics, including in-force premium per customer, gross loss ratio and net loss ratio, and a definition of each metric, why each is useful to investors, and how we use each to monitor and manage our business. In today's presentation, we hope to demonstrate to you three main points. The first, that Lemonade has a structural technological advantage that will manifest as superior loss costs and expense load over time. Secondly, that we have an in-built edge in acquiring and delighting new entrants, and their full value will unlock as we grow with them over time. Finally, that we have an advantaged long-term business model and a path to profitability. It's really precisely the structural technological advantage that begets us this in-built edge when acquiring new entrants.
It's the combination of both the structural advantage and the in-built edge that leads us to an advantaged long-term business model. This morning, we're joined today by our co-founders and co-CEO here in the room, Shai Wininger and Daniel Schreiber. Daniel will be kicking us off and sharing our long-term strategy for running our business and the technology that we're deploying. Maya Prosor will then follow, our Chief Business Officer, and give you our product by product runway and practical applications in how we run our businesses product by product. Finally, Tim Bixby, our Chief Financial Officer, will share the financials and some of the modeling work that we've been doing. We're gonna conclude, of course, with Q&A.
While each of the speakers are coming from different perspectives, it's really the combination of those three points that I just laid out that are a common thread throughout each of the presentations. I ask that you keep that in mind today as you listen in and tune in from around the world. With that, Daniel, the stage is yours.
70 years ago, Alan Turing, the inventor of the computer, already fathomed fully its implications, its ultimate implications. Writing in 1951, he spoke of how the forces that he had unleashed would inexorably lead to superhuman intelligence. In his own words, "It is customary to offer a grain of comfort in the form of the statement that some peculiarly human characteristic could never be imitated by a machine." He added, "I cannot offer such comfort." A lightning review of how AI has developed since the days of Turing until today and a snapshot of where we stand with a little bit of a propensity to look forward to where we're going will form a great backdrop for us addressing the question of how technology and AI will impact insurance writ large and the prospects of Lemonade in particular.
Well, AI, as I say, is a theoretical framework kicked off in the 1950s, and there were various starts and false starts and attempts and disappointments, winters by different names. In the 1990s, AI started becoming genuinely interesting 40 years after it began. Perhaps the seminal event is in 1997 when Kasparov loses to Deep Blue, IBM's supercomputer, three games to two, and suddenly AI has beat the best of humans at something that we thought was quintessentially human. Now, the AI that beat Kasparov was what we today refer to as good old-fashioned AI. It couldn't do anything that it wasn't explicitly programmed to do, and it bested Kasparov because of its brute force. In fact, it was grand masters who programmed the computer.
Now, unless you are a world champion at Jeopardy! or a grand master at chess, AI wasn't really a feature in your life in the 1990s, and it really wasn't until 10 years ago that AI started getting to levels of applicability that affected all of us. Then really big machines were met by big data, and suddenly old frameworks, theoretical frameworks like machine learning and deep learning revealed their true power. That today Google can translate from any language to any language without having been taught how to do that. Google Images can recognize me in some obscure photo far better than my mother can. When AlphaGo beat the world champion at the game of Go, it did it without any grandmasters teaching it anything.
It played against itself a billion games in a day, and by the end of the first day, it was better at Go than all of humanity's accrued wisdom over 5,000 years of playing the game. Pretty powerful stuff. Machine learning and deep learning, as they've come to be familiar to us, are fairly narrow in their scope. They do what they're being set out to do. They have specific tasks, and they perform them at a superhuman level. What's happened in the last, in theory, few years, but in practice only a few months, is what some people are referring to as a third generation of AI, or generative AI, as it seems to be emerging and being called. These are AIs that are trained on absolutely staggering data sets known as foundation models. What's amazing about them is that they have emergent intelligence.
They are proving themselves able to do things that nobody taught them how to do, that they weren't even intended to be able to do, with some fun consequences. I'll show you a couple of examples. I gave an instruction to one of these AIs. It's called GPT-3. I said, "Write a Shakespearean sonnet about AI and insurance." Now, bear in mind, this AI had never been told who Shakespeare is or what AI is or what insurance is or what a sonnet is. You give me till my dying day, and I couldn't produce anything like what it produced in three seconds. I'll read it out just for fun. "When I do think of insurance, 'tis but to ponder. How AI doth make it better. With nimble speed and keenest eye, the claims and the coverages do it spy.
In a trice with little fuss, the benefits and the best rates thus are found and given with but a click. Fair is insurance, I do thee praise for being made more excellent with AI. Now, maybe you fancy yourself as somebody who could write that. Maybe you don't. Let's switch gears for a second. A sister AI of GPT-3 was given this task: Draw an impressionist painting of insurance adjuster denying coverage after a fire. Now, maybe like me, you lack the skills to draw impressionist paintings. That's okay. Let's pit our power of imagination against the AI and take a second and in your mind's eye, paint what you would do if you were given this prompt. Well, this is what the AI produced. Pretty sophisticated, pretty subtle, pretty complex. If we're pitting imagination against AI, let's take it one step further.
I asked AI to render a photo of an insurance agent made out of a salad on a table. You wanna try that one in your mind's eye for a second? Pretty cool. Finally, though I could do this all day, I asked it to create a 1960s photo of a despondent insurance broker in a red dress sitting at a desk full of paperwork with fluorescent lighting, taken using a 50-millimeter lens, expired CineStill 800T film and an oversaturated filter. It came up with this haunting image of a woman in fluorescent lighting in the 1960s in a red dress, and she looks pretty despondent. Nobody told this AI what despondent means or what insurance means. The CineStill film that I put in the prompt was only invented a few years ago. This image never existed. This woman never existed.
It's pretty awesome what AI is able to do today. It's really against this backdrop, and with this kind of in the front of our minds, that I'd like us to start considering how Lemonade has approached the world of insurance. Because as we've stated from our founding day, Lemonade has been established as an AI doing insurance. That is the foundation upon which our company is built. Those are the fibers that run through our company. We gave ourselves a prompt a couple of years ago. I said, "We're gonna look at two pictures from our IPO prospectus." Well, draw a schematic of insurance as a neural network. This is one that we produced, not the AI. The picture that formed part of our prospectus those two and a half years ago was making the following point. Lemonade has but one brain.
It is vertically integrated. All of its systems are interconnected in powerful ways. The same intelligence that sells policies, handles claims, answers customer inquiries, identifies fraud, it's all one centralized intelligence. I wanna put it to you as we go through the next few minutes together, that this kind of AI, this kind of platform, this kind of insurance company could mark the dawning of a new age in insurance. Hyperbolic as that may sound, I'd like to try and make the case for that. The closest analog that I could find to what I think may be going through, what insurance may be going through today happened a while ago in the scientific revolution in the seventeenth century. In the seventeenth century, a lot happened in the world of statistics. Pascal and Fermat framed the basics of probability theory.
Jacob Bernoulli formulated the famous law of large numbers, and suddenly statistics was put on a firm footing in a way that it never had hitherto. Now, insurance is fundamentally at its core, the business of monetizing statistics and data, and it would therefore be no surprise if there was a discontinuous innovation in the field of data and statistics for that to expose a rift, to create a disruption in the world of insurance and for it to advantage people and companies formed post that rift. It was. Not a single insurance company, and they've been around for many, many, many millennia.
Not a single insurance company that predated Pascal and Fermat survived the scientific revolution. Whereas the newly founded companies in the late 1600 and early 1700s that use these newfangled ideas about how to handle statistics and data, well, they have not only survived, but they have thrived for these 300 years. Let us not confuse longevity for immortality, because what we are living through now, the digital revolution, is akin to the scientific revolution in the changes that it is creating in the world of statistics and data. Let us dive in a second and see what this schematic looks like in the real world. 2.5 years ago, it was conceptual. Let me show it to you in action. Beforehand, two data points as we consider what I'm about to show you.
One is just a sense of the dimensionality of what we've built. We went live just six years ago. We're a young company, and as you'll soon see, our data and statistical capabilities, our machine learning has been building at an exponential rate, but it is still in the early innings. Nevertheless, there's a lot to look at today. We now have several hundreds of millions of customer interactions in that centralized intelligence. That intelligence has digested and analyzed some 160 TB of pretty textured and predictive data. That's a lot. All of that has resulted to date in about 50 machine learning models which are at first approximation AIs unto themselves. We have different AIs trained to do different things using these datasets. While the sheer magnitude of the data is impressive and important, it actually misses the most important point.
It is not the tonnage of data that matters, it's the interconnectedness. This is where traditional insurance companies falter. Systems built higgledy-piggledy over years, and acquisitions in decades, in different geographies that are not connected are very poor at doing what we're talking about. You have data lakes that are a mile wide but an inch deep. By the way, talking about intelligence, this is how the human brain works, right? We have something like 100 billion neurons in our brain, but intelligence doesn't lie in the neurons, but in the connections that they make. Each neuron has some 10,000 synapses, which are the connections, and therein lies intelligence. This is where the traditional model falters. From the schematic, I'd like to shift gears and give you a live demo of some real stuff.
On the screen now, what you're seeing is a live feed of a visualization of Lemonade's actual Customer Cortex and its systems. I'll talk you through what we're seeing. It is beautiful, but it is also a little bit confusing at first. I can manipulate this, and we'll zoom in and out and enjoy ourselves as we go about this. But let me try and show you a little bit of what we're looking at. I'll wipe them out and I'll go through the different colors and explain myself. At its core, you can see several pink balls, relatively big, and these are applications. Some of these will be more familiar, some will become familiar as we go through the day.
For example, AI Maya, if any of you have met Lemonade before or played with our bot, you'll know Maya is the bot that sells policies and insurance. What you won't have known is that she calls on many, many machine learning models, and that's the next set of balls that you're seeing there. All these blue balls are those 50 machine learning models or so that I spoke about a minute earlier. Each of those machine learning models, in turn, is fed by hundreds of features. Features are data points that the machine learning has found to be highly predictive out of a plethora of potential data points. The yellow balls each represent a dataset. I mentioned 160 TB of data, hundreds of millions of interactions.
You're only seeing a few yellow dots because each one contains within it millions or tens of millions or hundreds of millions, sometimes billions of data points. We've got rid of all the data and just showing you the structures so that it doesn't get overwhelming. One of the points I really want to drive home now is if we stay on AI Maya here for a minute, if I can find her dot there. You can see she relies in real-time. This isn't stuff that is cached. This is every time she sells a policy, this happens again. She's calling on all these different machine learning models. If I click on her, you'll see all the connections within the cortex, within this system that happen as Maya lights up and does what she does. There's another model I'd like to spend a minute on here.
Let me darken out Maya for a second. This is the LTV model. You'll hear a fair bit about this during the course of the day. Our LTV model is a series of AIs. You can see all the blue dots there that are lit up, and a stunning amount of data and other stuff behind the scenes. You can see they're lit up now. Zoom out a bit and you can get a vantage point. You can see again that a big chunk of our brain lights up. In fact, once we light up Maya as well, you see that between them, they all are interconnected in powerful ways. At any rate, staying with LTV for a second. The LTV model is a series of AIs making real-time predictions. Every single prospect, every single customer that comes into Lemonade, these machine learning models light up.
The whole system here lights up and makes a series of predictions. The ones that I wanna highlight to you are it will make a prediction about how likely a customer is to churn. In other words, talk about lifetime value. What is gonna be their lifetime? Are they gonna be here for a month or for 50 years? How likely are they to claim, and what will be the severity of their claim? And how likely are they to cross-sell and buy another policy? These are some of the most important variables in our business. Those three Cs, other things as well. If you think about it, once I'm able to predict those three, and I know what I'm paying to get the customer, the CAC, the customer acquisition cost, I can start making highly pointed decisions about how to allocate resources.
Because what LTV is really doing is it's showing me, and it will produce a dollar number at the end of the day, which is all the dollars I'm gonna spend on this customer, paying claims and otherwise, all the dollars I can expect from this customer in premiums. Apply a discount for the time value of money and produce a dollar amount and say this customer's lifetime value collapsed to today is $1,300, $5,000, -$2,000. Insurance is a business for some customers you don't want for free. We now run our business in tremendous reliance on increasingly using these LTV models.
We've seen a little bit about how all these systems interconnect, and I put it to you that there isn't another insurance company in the U.S., for sure, in the world most probably, who has something akin to what I'm just showing you right now. Let's try and look at some of the applications and implications of what I'm showing you. Again, for the first time, sharing here real-time data from our systems. This is connected to the web, which is why I'm on a different computer, showing you a real feed from our business. Although we're starting at a time-lagged spot. If you look at the bottom of the screen, you'll see the dot there is on Q4 2019. This is the very first LTV model that we had three years ago. We're now in Q4 2022.
Every dot here is a homeowner's policy, so we're just gonna look at homeowners right now. As it stood in 2019, and this is the machine's estimate of the value of each of those policies. Color-coded, light green is marginally positive, dark green is very positive. As it gets into the red, it's negative. A couple of things as I step through this. By the way, at this stage, we had so few policies. Think about it, homeowners make claims seldom. We had many fewer claims data. The machine couldn't yet make predictions based on claims. It was using proxies. We could use churn, and we're using churn as some kind of a proxy. This was not yet a reliable model, and we knew it, and we didn't use it yet in production.
As we step through, you'll see that, A, the dots become more numerous because our business is fast-growing. Hopefully, you'll start seeing some separation. The important thing to see with machine learning models isn't whether the business is getting better or worse. That's to do with the business. If the machine is getting better or worse, you'll see by its ability to discern differences, more lift, more nuance, less monolithic, more precise. The real breakthrough for us was in LTV 4. This is just 18 months ago. This was the first time that the machine said it had enough claims data to start making the kind of predictions like I shared with you. It's the first time that we allowed it into production. Until that point, it was still just learning and being trained. 18 months ago, we started using it.
By the way, at a glance, you can see a lot about our homeowners business. Those of you immersed in insurance will be less surprised perhaps than others, but California is looking pretty red here. Texas, bright green. The Northeast, a mixed bunch. New York, pretty good. Virginia, less good. I'll click through a little bit more. You can see LTV 5, more separation, more nuance, more color. LTV 6, which is our most recent model. Again, things are getting better, but beyond the fact that they're getting better, what I wanna draw your attention to is the fact that they're getting more nuanced. Every time monolithic groups break apart and reveal their subgroups. Let's zoom in. Let's go back for a second to a place that was looking bright green, which is down here in Texas. Let's go to Houston.
I won't get so close that you can actually tell individual addresses, although we could. LTV 1, you can see the sparsity of data. As we step through, more and more people in Houston are buying Lemonade. By LTV 3, it's looking pretty good. Then LTV 4, the predictions say actually we were underestimating the value of our customers there. Everything turns dark green. The machine's saying, "Houston, we have liftoff. Everything's looking really good." Now, that doesn't change at a zoomed-out level. Houston continues to look good. Here's the power of the machine as it goes through its iterations. I go through generation 5. You start seeing some more shading. In generation six, yeah, at a zoomed-out level, everything's green, but you can start saying that not everything is equally green. In fact, down here there's a patch of red.
Friendswood isn't looking so good. We are mispriced in Friendswood. Now, we couldn't tell that until very recently, but we know that now. Fresno, just 15 miles out to the west, is looking bright green. We can start analyzing what's going on in these two markets. Very similar populations, very similar range. Something's wrong with our pricing, and we can go in and fix that. Now, stepping through that map hopefully gives you a really powerful sense of how the machine learns and improves and changes, and in some ways, how recent all these capabilities are. If I try and quantify it for you a bit, this exponential pace of innovation. A couple of years ago, those features that I spoke about that the machine learning models call upon, we did about 250 million calls to them in a year.
This year we think it's gonna be over three billion calls, so more than 10x in the space of a year. This is important. How many predictions did all of these different machine learning models, our totality of this intelligence that we built, how many predictions was it able to generate in a year in a way that helps our business? Two years ago, 8.5 million predictions. That sounds pretty impressive, but it pales into insignificance next to the 110 million that are happening this year alone. Why do we care about all of this? It's cool. We like cool stuff, but what are its business implications? Why is this really, really, really important? Why might this herald the kind of change in insurance, a dawning of a new era that I was speaking about earlier?
Well, because these kind of capabilities result in more for less. More is for the customer, more delight. Business Insider did a review of Lemonade, and they concluded that the cheerless customer service representative at Liberty Mutual had about half the personality of Lemonade's chatbot. Now, we've seen that AIs are able to produce beautiful pictures of insurance agents out of salad on a table. It's not shocking that they can also sell insurance in a trice. Pretty much any way that you care to measure customer satisfaction, whatever your word salad is, whether it's NPS or CSAT or BBB or whatever, Lemonade is performing at a level on par with companies like Tesla and Apple and really on a level that is unfamiliar in the insurance space.
Well, it's becoming more familiar in the sense that the mainstream arbiters of customer satisfaction have woken up and paid attention. Now J.D. Power and Forbes and U.S. News & World Report, all these guys are now ranking Lemonade when they're running comparisons at the very top of their rankings. I'm not gonna delay too much on this. Hopefully, you don't need a lot of convincing that in terms of customer satisfaction, our AI and our system is outperforming incumbents by some margin. Let's talk about the loss part. One of the central themes that we wanna convey to you today is our conviction that the system that we have built is gonna show savings, crush costs in every single line in the P&L. Every single part of our combined ratio, and I'll use that to talk you through it.
What you're seeing here is the expense ratio, the loss ratio. Unimaginatively, they are called the combined ratio. Let's step through a few ways in which this intelligence is affecting all of these. Some are invisible to the outside world. For example, we have an AI internally, a bot called Cooper, and Cooper runs loads of errands for us, loads of tasks that are usually done by humans. Pretty sophisticated tasks. He runs big chunks of our engineering. While he doesn't write code at the moment, he does do what's known as DevOps, which is establishing server environments, instantiating environments, giving them to developers, moving things into production. Last year alone, Cooper pushed about 12,000 production builds of software into production. 12,000. He saved us, we reckon, something like 10,000 human hours by doing that.
At the other end of the spectrum, you have Maya, with whom we are familiar. AI Maya and her associated APIs sell all of our insurance, 98%, and she does that in a trice. 90 seconds and you're done. No hassle, no commission. 95% of homeowners' policies in America, give or take, are sold by humans, and they command something like a 15%-20% commission, not just for the year they sell, but in perpetuity, and not just the policy they sell, but on all subsequent policies that the customer buys. Maya is much more forgiving in that sense. We spoke a fair bit about the LTV model. Our conviction in trusting this model, we're putting our money where our mouth is, quite literally. 86 cents on the dollar that we spent on marketing this past quarter was at the direction of the LTV model.
The graph on the left shows two overlapping graphs. One is showing what the model says would be optimal, one is showing what we actually did. You can see that they're almost entirely on top of each other. The bars on the right being so clumped together gives you a high conviction that all the scenarios that it's running, which are all of those little lines, are coming out to the same conclusion, so very high trust level. A big chunk of our customer engagement, post-acquisition and pre-acquisition, is done without any human involvement. We have an intelligence called CX.AI. The users actually interact with Maya, but what we call it behind the scenes is CX.AI. Nowadays, a third of all customer inquiries, no matter what they're asking. Some of these things are pretty complex. You know, you're writing in that you're moving home.
Well, we have to figure out when are you moving and when do you want your policy canceled, and then what coverages do you need on the new home and which date and all of these kind of things, and which scheduled items do you wanna add and which family members, and all that's done by the AI without any human intervention. A third of all of our inquiries, one in three, is handled this way with an incredibly high level of customer satisfaction. Loss adjustment expense, as we switch to loss ratio for a second. Loss adjustment expense is a big part of the loss ratio. That is the bureaucratic overhead of managing claims. 98% of our claims, the first notice of loss is taken by a bot.
In almost half of those claims, everything is done by a bot, start to finish, right through to asking you any clarifying questions, asking you for documents, you uploading a video, analyzing fraud detection, asking you all the questions and wiring money to your bank account, all done without any human intervention. By the way, as I write it's now the level of accuracy we audit this. AI Jim is performing better than humans in most of these tasks. Again, big chunk of loss ratio is fraud. It should come as no shock that the kind of intelligence that I've described, the kind of AI, is getting pretty good at detecting any kind of deviant behaviors. What you're seeing here on the left is a graph.
The machine learning is scouring through all these data to look for connections to suspicious behaviors, something that, given all the time in the world, humans could never find. It's just much too much data to handle that way. Finding these connections in this graph is actually a graph of these connections. We now have nine million connections that we've found within our data sets of suspicious behavior. This, as you can see, is growing at an exponential rate, which means we're getting better and better and better, more and more and more precise at identifying these kinds of threatening behaviors. The financial impact is profound. These systems to date have identified about $100 million worth of fraudulent claims that we then inspected with humans and found to be fraudulent. It flagged $12 million worth of claims just from the fraudulent documents.
We are now getting pretty good at identifying the fingerprints on a document of it being doctored in any way. You'd be shocked at the kind of things we see. But humans miss these things, AIs don't. We have something that we call Watchtower. Watchtower is a system that gets satellite feeds in real time from NASA to look for catastrophic events. It gets other data from sources as well. If it knows of something happening, it'll take preventative measures. A pretty terrible hurricane hit Florida two weeks ago, Ian. As Ian was approaching Florida, 6,400 Floridians turned to Lemonade for insurance. Watchtower kicked in and offered all of those people a policy so long as the start date was 10 days forward. 31% of them bought.
You get all the benefits of the fact that catastrophes do remind people that they need insurance without the associated risks. We're gonna delay on the next few slides in Maya's presentation, so I'll fly through them. Telematics are a game changer. Profound, and in use at Lemonade at a scale that is simply being missed by the industry at large. In our homeowners business, we are using computer vision and natural language processing to identify incredible signals about the state of properties, and we'll delve into that more in a few minutes as well. Finally, and this too we'll come back to, for the very first time, we've now had a machine learning model adopted by regulators as the foundation of pricing for homeowners. What you're seeing here on the image is what's known as a boosted tree.
This is a form of machine learning, and each of those threads is a different path in which customers might go, and you end up with incredibly precise and nuanced outcomes at a level far beyond anything that traditional filings can get to. This calls on the machine learning model in real time. Okay, there are a few buts that you might have. I have none. I'm gonna try and anticipate them. The first one is, "But everybody's doing this." Cool, but this is now stuff that surely the incumbents, surely everybody can do what you're showing. My monosyllabic answer to that is no. I'll elaborate. Anybody can pay a third party for a suite of software. Anybody can pay a consultant to run some analysis. What we're showing you today is different, not in degree, but in kind.
The kind of connections, the kind of systems upon which we are built, nobody else can do. Unless you were built this way from the ground up, you just can't make the connections, and it's through no fault of anybody else. If you founded your company in the era of the horse-drawn carriage, you optimized for that era, and you find yourself flat-footed going into the digital era. We have the good fortune of being founded in the digital era, and therefore, we built it the way we built it. No, I say this with some confidence in having spoken to many of my colleagues, CEOs of large incumbents. Other companies cannot do this. In fact, what I'm showing you is a sustainable competitive advantage that is already manifest if you look at the right places and will manifest ever more powerfully over time.
I hope you take my word for it, but just in case you're suspicious, I want to share with you the viewpoints of three people, all members of Lemonade's team, who joined Lemonade from incumbents, from senior positions in some of the best and brightest in the industry. Let's hear in their own voice how they compare their experiences at those companies and at Lemonade.
Hi, I'm Scott Fischer. I'm Lemonade's General Counsel.
I'm John Peters. I'm the Chief Underwriting Officer for Lemonade.
I'm Sean Burgess, Lemonade's Chief Claims Officer.
A few years before I came over to Lemonade, I was the insurance regulator for the State of New York.
Prior to Lemonade, I was a leader in McKinsey's global underwriting practice, and I was also a chief underwriting officer at Liberty Mutual.
Previously, I was Chief Claims Officer at USAA, a best-in-class company. I can say without hesitation that the tools and technology that we're deploying here at Lemonade are simply unmatched in the industry.
I don't know of another insurance company that has the capabilities that Lemonade has to use AI.
I can honestly say I don't think there's another company with the vision or the DNA on innovation that Lemonade has. Its people, its tech, its general approach are second to none in the industry.
I think one of the powers of having built our technology from scratch and literally from end to end, from the moment a customer hits our website to the time we pay a claim, all of that's integrated and connected together.
Those facts are a material competitive advantage in an otherwise really challenging marketplace.
In fact, I don't see a scenario where incumbents will be able to get here from where they are today. It's a true structural advantage to Lemonade, and the best news is we're just getting started here.
Why aren't we seeing it in the financials? Lemonade is losing money and has a high loss ratio. How do you square that with everything that we've just seen? I'd like to make three points and try and explain that disconnect. The first is that we are seeing it in the numbers. It depends which numbers we look at. We had a 94% loss ratio in Q3, but based on all the data we have, the business that we wrote in Q3 will have a lifetime loss ratio of 61%. The first point to understand is, are we looking at lagging indicators, which are the financials that we report quarterly or leading indicators? In a fast-growing business, you just saw literally on a map how much has changed in the last couple of years for Lemonade.
So much of our current financials is a result of what happened a year, two, and three ago. Look at leading indicators rather than lagging indicators, and you'll get a diametrically opposite picture. I wanna dwell on this one for a second. I spoke about the LTV model. Part of that model, it produces LTV, but on its way it produces an expected lifetime loss ratio. Lifetime loss ratios are important because loss ratios tend to be front heavy. Customers have the most claims, the worst loss ratio in their first year, gets a bit better in the second year. In our experience, they tend to be around their lifetime average loss ratio in the third year, and then they go below it in years four and five. When you're selling business, you wanna think about long-term profitability of the business.
You don't care if they are loss-making the first few months. You wanna think about will they be loss-making in a meaningful way over their lifetime. You wanna be forward-looking, not looking at a snapshot right now. The machine learning model does all of that. What it is showing you on the screen here right now is that in Q1 of 2021, the cohort that we onboarded, not the totality of our business, the business that we sold in Q1 2021 on aggregate across all of our products, the machine forecasted a lifetime loss ratio of 86%. You can see how quarter- on- quarter- on- quarter it has dropped precipitously to the low 60s to 61% in this past quarter.
There's nothing at all inconsistent with saying that our actual reported lagging loss ratio was one number, and the new business that we were writing was something quite different. The loss ratio of the people that we sold policies in Q3 didn't impact our Q3 loss ratio almost at all. Now, I don't wanna vouch for the decimal point here. Actually, I'd prefer to white out the numbers entirely. What I will vouch for is the trend line, because this is an apples to apples comparison. This is using the same methodology to evaluate the lifetime loss ratio. You white out the numbers, and you can see that quarter-over-quarter, we are bringing in better and better and better business. To fully understand why there is such a difference between leading and lagging, I wanna make my second point.
Insurance in general, but Lemonade specifically, is front loaded in its expense, in its losses. The near-term financials do not provide a helpful metric, a helpful signal for long-term profitability. There are four ways in which we're front loaded. The first is that as a company that is building all of its own technology in-house, that is getting all of its licenses itself, writing on its own paper, vertically integrated to its core, as you launch more and more products, and we've launched a lot of them over the course of the last six years, you are incurring all of those costs up front, all of the development, all of the engineering, all of the regulatory work. Excuse me. You have all the building costs up front. You start off with a data disadvantage.
You saw that beautifully on those maps and how sparse the data was and how long it takes, years for those cycles to run through and for the data to accumulate. You have what's known as the new business penalty. Insurance insiders know this, that new policies, new business always performs worse than it will over time. Guess what? When you're new, all of your business has a new business penalty. For incumbents, maybe 10% of their business is new. For us, it's like 70% or 80%. Finally, I made this point fleetingly earlier, we are a direct-to-consumer company, which means that we incur all of the customer acquisition costs on day one. Broker-based businesses, they don't have that. There's no one-time expense up front. They have a partner taking a commission in perpetuity.
You add up all of those dollars, it may be much more than what we're paying, but it's spread out. It's peanut buttered over the life of the customer. It's not front-loaded in the way that it is for us. Here's the good news. All of these resolve themselves over time. When you're finished building products, you sell them, and you start recuperating the investment that you made. Data disadvantages, as I hope I've demonstrated, at Lemonade, turn into data advantages pretty quickly. I will tell you that today, six years in market, there is not an insurance company in the U.S. that we would trade datasets with. New business penalty? Well, business seasons all on its own. If you're growing super fast, you'll still have loads of new business, but you know where it's going.
Finally, the beauty of acquiring customers through CAC is that year two, you're not paying anybody anything. You end up with no commission sales for the rest of the life of the customer. Which brings me to my third point. Lemonade is built for scale. Our ambitions are expansive. I just spoke about how we're getting all of our licenses, launching our own new regions, building our own technologies. You don't do that if you wanna stay at the size that we're at today. You do that because you're envisaging many, many years of profound growth, and you're preparing the infrastructure for the company you intend to be, not for the company you are today. That means that you have an expense load that is disproportionate in the early days of the business, but that's the right thing to do if you're thinking about the long term.
Using this image, you can see the denominator is what matters. The expense load isn't the problem. The problem is that we still have a small denominator, and an expense ratio is all determined by the denominator. The numerator matters, of course. You will see our expense ratio resolve itself as our business grows. In fact, it already is. Let us look over the last few quarters, Q3 to Q3. This is our adjusted expense ratio. Expense ratio of Q3 last year was 103%. This year was 80%, and the trend line, I think, speaks for itself. This happened while our business grew at 76% during that same period. It's no coincidence. The growth of the denominator is what resolves the ratio all by itself. We are on a path to having a fabulous expense ratio.
It's just gonna take time for the business to grow into all the stuff that we have built. All of that brings me to the second image in our prospectus that I wanted to talk you through this morning. The prompt here is growing with customers. This is the image. This was a picture that we included. It's a picture of a young woman at age 25, starting out on life, in life. She has a bike. She has a backpack. She has maybe a laptop in it. That's what she needs to insure. We offer our policy at $60 a year, $50 a month. She joins us, she gets a renter's policy, she insures the few things that she cares about, and then she sets off.
She steps onto the conveyor belt of life, and if we've acquired the customers that we think we've acquired, she's gonna go through pretty stereotypical, predictable lifecycle events. It's a bit of a caricature, but you know how this ends. It ends with her going from $60 a year to $600 a year to $6,000 a year. We made it very clear that we plan to be there with her for life. In general, when you talk about lifetime value in tech, you typically think in terms of three years, four years. In insurance, it's literally for life. Age 80, she will still be paying insurance to somebody, and she will be paying perhaps 100 times more than she pays when she starts off. Now, when we threw this graph up on the prospectus, we were a monoline business.
2 . 5 years ago, we had but homeowners, renters. Since then, we've filled in pretty much all of that hill. Renters is still in this story of starting with the young consumer, this low-end disruption of finding consumers that incumbents want the least delight in them and then growing with them was the intention all along, as I think this graph bespeaks clearly. How are we doing with that? Last month, Google ran a survey of asking Americans across the nation if they bought renters insurance for the first time in the last 12 months, and if they did, who they bought it from. You can see at a glance that State Farm came in tops. Number one brand that people bought renters insurance for the first time last year across the USA was State Farm. Brand number two, Lemonade.
Ahead of GEICO, ahead of Progressive, ahead of Liberty Mutual, ahead of everybody else. By the way, we were handicapped in this survey because we're not live in all of the U.S., and the survey was conducted across the U.S., but let that be. When you then splice the data one level further and you say, "Okay, show me those same results, but just for under 35-year-olds," Lemonade bests State Farm and becomes the number one brand in the nation for Americans buying insurance for the first time. I put it to you that this is powerful and strategic, that nothing foretells ultimate market share more than new cohort market share. Now, it's not enough to just acquire these customers. We then need to grow with them. The second part of that sentence. Acquire them when incumbents want them least and then grow with them.
Well, at IPO a couple of years ago, we were doing that already, but we were limited. The only growth they could do with us is when they're ready to get rid of their rental and move into a condo, typically or homeowners policy. We tracked that pretty closely. At the time of IPO, 11% of our condo business was sold to existing renters. Today, two years later, that has steadily improved. It's twice that today. More than 20% of our condo business was sold to existing customers. Now think about that. You acquire renters at the cost of renter acquisition. It's profitable business in and of itself. They grow with you, typically expanding their premium about 5x, with no associated cost to acquire the incremental premium.
We've moved a long way in those two and a half years, 'cause we've since then launched Pet and Life and Car. Folks, look at these numbers. This past quarter, roughly 1/3 of all car policies that we sold, roughly 1/3 of all pet policies, roughly 1/3 of all Life policies, and roughly 1/3 of all Home policies were sold to our renters. This has the making of a very, very powerful business. I also wanna tell you just how early we are in this process. Let's put some dimensionality around this as well. For incumbents, the best and the brightest, something like, they don't publish these numbers, but as best we can gather, something like 60% of their customers are multi-line customers. They've gone through this. They've bought more than one policy. Not just multi-policy, but multi-line.
At Lemonade, it's less than 4%. If that sounds like bad news, it's fantastic news. It means that we have tremendous opportunity ahead of us. We're on track. Let's look at some of the trend lines in this regard, just looking at the multiline customers. 2.5 years ago, we had zero. We didn't have multilines, so zero customers. July 2020, we launched pet, and then we started our journey in the multi-line business, followed by Life, and there you get to the 3.7% that we're at today. A nice up and to the right trajectory. Then something happened. One year ago, almost to the day, we launched Car. That was in Illinois. We've launched it subsequently in other markets, and the trend lines are identical.
Look at the angle of the curve in Illinois, where we have car available as well, and extrapolate that forward. Illinois, a year later, is already at 5.9% and shooting right up through the screen. I wanna wrap up my presentation and say the following: these two images that I've used as an anchor, as a visual aid to talk you through the building blocks of our strategy, when we drew them out 2 . 5, three years ago, they were ambitions. They were a statement to our investor community of what it is that we intend to do. Today, they are a reality. I've shown you the data supporting this young woman's journey. I've shown you literally our back-end systems and how they reflect the schematic that we drew those three years ago.
I put it to you, and as we go through the morning, we should return to this point again and again, that they represent a highly differentiated business model from what the rest of the industry does. Neither of these graphs could have been really presented to you by anybody else. They are highly distinct, highly differentiated, and I put it to you, highly defensible because they are structural in nature. Now, we're just getting started. We have built the flywheel, and as I hope I've shown you, it's beginning to spin pretty quickly and accelerating. I'd like to tell you where at least we hope it's taking us, and this will be my final slide. We're tiny. This is drawn to scale. GEICO towers over us 50 - 1, more than. State Farm, more than 100 - 1. Others, even more.
We find it exhilarating that we could 10x our business and 10x it again, and State Farm would still be bigger than us. There's just that much headroom. That flywheel spinning, that's exactly where it's intended to take us to go. With every spin, our business not only gets bigger, it gets better, smarter. It learns from the data it generates, and that moat that I spoke of gets taller and thicker. To show how all of this impacts our business day-to-day and line by line, let me hand over to Maya, our Chief Business Officer.
It's great to be here. We've done a lot of great things, but we're just getting started. We spent the past six years building five incredible products. Each of these products represents a total addressable market of billions of dollars, if not hundreds of billions of dollars. My job at Lemonade is to manage our portfolio, and so it's great that we now have such a rich portfolio. That's pretty recent. Up until 2020, we were a monoline carrier offering insurance in just 28 states. Today, we have all major personal lines. We have customers in 50 states and even Europe's three largest markets. The question we ask ourselves every day is, where should we be spending that incremental dollar? Which product, what geography, which campaign would yield the best return to our investment?
Not being able to look into the future, insurance companies oftentimes relied on information from the past. As the warning goes, past performance does not guarantee future results. Take our loss ratio, for example. It tells you a lot about the business we sold two years ago, but it doesn't tell you a lot about the business and policies we sold just two days ago. You're investors, but if I told you that you need to manage your own portfolio today, deciding which stocks to buy and sell, using only the information available from The Wall Street Journal of January 2021. At Lemonade, we do things differently.
We report lagging indicators, of course, but we use leading indicators to manage our own business. Our LTV model looks at every policy we sell and predicts the possibility of that policy to churn, the predicted loss ratio, both in terms of frequency and severity, and even the potential of that customer to buy other products from us. Like with any probabilistic model, when you look at it one customer at a time, the model might be off. Aggregated together, it provides a reliable and powerful set of tools, and this is how we manage our business. Let me show you how that looks like product by product.
I'm gonna get through Renters, Pet, and Life fairly quickly, and then I'm gonna spend a little bit more time on how we deal with the challenges that we have in Home and Car as they represent the biggest potential for our growth in the future. We launched Renters six years ago, and the short gist of it's doing great. It's profitable. We've been growing this product for the past six years consistently, and according to Google, we now have 9% market share. We've sold $223 million of IFP, which is exciting, but compared to our aspirations, the more exciting number is 1.4 million customers. As Daniel talked about, these are customers who are first-time insurance buyers who've never purchased insurance before, and these are future homeowners insurance buyers, Life Insurance buyers, Car Insurance buyers.
All of them are going to graduate into those products with zero additional cost of acquisition. We haven't even shifted our focus there because many of these products are new to us, and what we're seeing is a lot of these renters are buying all of these products all on their own. This graph represents the contribution of IFP just from Renters to our other product lines in the last few quarters. In Q3 of 2022, if you can squint, you can see $3 million of IFP coming to our Car business, which is our newest product, just from Renters. Mind you, we're available in just three states. The most bundled product for Renters in the U.S. is Car Insurance. We expect that line to overtake all other products and contributions.
In Q3 of 2022, $25 million of our IFP for our Pet Insurance business came just from Renters. Speaking of Pet, we think about Pet Insurance very similarly to how we think about Renters. These are younger customers, many of them are first-time buyers of insurance. What's more interesting is that this really creates a different relationship between us and our customers. It's moving us from insuring the stuff you love to insuring the ones you love. Our customers look at their pets as part of their family. Every time we engage with them and we pay out a claim, we create a deeper connection with them, which creates stickiness and more opportunities for us to cross-sell them and offer them other products. We're excited about pet insurance also because of its market, which is essentially a blue ocean.
Only 2.5% of households with pets have Pet Insurance. There's a lot of room to grow. One of the reasons why pet owners don't buy Pet is because of awareness and education. We've built a phenomenal product that has been able to be broken apart, and so every customer can buy exactly what they need for their pet and understand the value of what they're buying. This market is growing 25% year- over- year, and we're growing four times faster. In the past year alone, we've doubled our Pet business. We just announced a strategic partnership with Chewy, offering Lemonade Pet Insurance to its 20 million online customers who are already used to buying Pet Insurance online. The numbers show this.
Numbers for Pet Insurance are looking really great, but the number that I wanna focus on is might be less obvious, and this is the number of claims. We've had 450,000 Pet claims in the past two years. The reason why we love Pet Insurance and why this number is great for us, it's because Pet is one of those products that has a really low severity and high frequency. This is a product that is meant to be used. You're meant to be taking your pet to the vet a few times a year. This is how we price the product. Every time you do that, we learn more about you and your pet, and we're able to price and underwrite these customers better. Let me show you how our Pet claims look as opposed or comparison to all of our other products.
The pink line is all of our Pet claims, and the gray is all of our other products combined. It took us seven months to reach the same number of claims we had in Home and Pet. Just think about that volume. It helps us with loss ratio, it helps us with pricing, it helps us with underwriting, but it also helps us, might not be as intuitive, with the cost of handling claims. The more claims we get, the more opportunities we identify to inject technology and automation in the claims handling. When we just launched Pet, 36% of our premium went to the cost of handling claims. We're already down to 14% with a clear path by the end of next year to reach below 10.
Life Insurance is one of our more profitable products, but it's small. We haven't managed to crack the cost of acquisition in the current market, and so we've shifted a lot of our resources and marketing spend into other products. This is part of the power of what I mentioned at the beginning of having a portfolio. We focus mainly at cross-selling this product to our other customers, and so it's not surprising that 35% of our Life customers have other products. The rest are coming to us organically, and it's reducing churn for the customers who are adding it. We're gonna continue to monitor market conditions and push the marketing into this product when things change. Homeowners insurance is one of the cornerstones of personal lines. This is the product that our 1.4 million renters are going to graduate into.
This is the product you stay with for many, many years, and the product that you bundle and add other products to. Very, almost mirror opposite to what I shared with you about pets in terms of the claims experience, it's mirror opposite. This is a product that has very low frequency and very high severity. For our datasets to be able to learn and be able to price and underwrite these customers, it just takes longer. You can see that in our numbers. Our Q3 loss ratio was 119%. Highly inflationary year, some catastrophic events, not where we want it to be. What I want to focus on and highlight is the gap between our ultimate loss ratio and our predicted loss ratio for the new cohorts for homeowners.
More so, the trajectory of how far we've come and how much better we've become in underwriting and pricing new homeowners coming into our book. The way we use predicted loss ratio, and this is important, is in two ways. As Daniel mentioned, we wanna look at the trend. We want to make sure that we're getting better and smarter with every new cohort that we're bringing in. Also, we want to find correlations between initiatives and decisions that we're making with that trend line. You wanna see that impacting new cohorts coming in. This is a very powerful tool to manage a business today versus managing your data from the past. A lot of the things that we've done are gonna take time to see them in our ultimate result. Rates are a really good example.
This past year, we've rerated 100% of our book and then started over again. Only 85% of the rates that we've submitted have been approved and implemented into the markets. But just 4.5% of it was still earned. It takes time for all your customers to get through the renewal process and then month-over-month, start accruing the premiums that they're paying us. Another benefit that we're gonna see in the future comes from our LTV. Our LTV doesn't know that more rates are coming. It just knows about the rates that were implemented. We expect the predicted loss ratio to continue to go down as the rates get approved and implemented into the different markets.
I want to share with you a little bit of what we've done in just the past 12 months when it comes to underwriting and pricing for our customers for homeowners. When it comes to homeowners insurance, geography matters. California has been a tough state for insurance companies. We've seen a lot of insurance companies dropping business and discontinuing business in California. The combination of wildfires and a tough regulatory environment that doesn't allow you to raise rates as fast as you would like has really created a challenging condition. In the beginning of 2022, 30% of our new homeowners business was coming from California. Not surprising, it's a large state, and as other carriers are leaving, you're attracting more customers to Lemonade. These customers were mispriced.
By utilizing and leaning on the fact that we're a direct carrier, we're able to shift marketing dollars as well as change some of our product flows and request more data from customers and lower that percentage all the way down to 6%. Today, we discontinued our direct business in California. It's pretty recent. Because we're not in the business of selling unprofitable business. We will turn that back on once our rates get approved, and we can rate our customers appropriately. Let me show you some of the things we've done in the past 12 months around underwriting. We've made a huge leap in the types of data and the datasets that we're using in order to underwrite our customers. To orient you, as Daniel showed you a little bit the Cortex, you should be familiar with this already.
This blue dot is really where the model that decides on the different underwriting flows and questions and data sources that feed us and make a decision on whether or not we should be underwriting, we should be binding that customer. We use many data sources. Some of them include the age of the house, the age of your boiler, did you do any renovations in the past year? It's how we use these data sources that really differentiates us because this feeds not only our underwriting review team, but this also feeds into our LTV model that helps us make marketing decisions. Every dollar we spend, we know that we're making it on customers that are eventually gonna get through our underwriting filters. Making sure that we have a lot more efficiency in how we spend money. We don't stop there.
We also look for additional sources of information that may not be as obvious, but definitely make sense. How customers describe their homes on real estate websites is pretty telling. This home, for example, says that this home needs TLC inside. Our models in AI identify that the word TLC is usually correlated with disrepair. Disrepair is outside of our underwriting guidelines. We're able to flag this automatically to our underwriting review team, make sure they don't miss anything, but also make sure that we're underwriting the right policies. Inspection reports are a really useful tool for us to know the condition and everything you need to know about the home. I don't know how many of you have ever read an inspection report. It can be as long as 120 pages.
On average, it's 60-70 pages of very complicated language, but really has everything you need to know. More than 60% of our homeowners are first-time home buyers. For them, this inspection report and for us is really telling about the condition and what's happening in the house. We do not want our underwriting associates to be reading through all these inspection reports all day long. We have thousands and thousands of homeowner quotes a day. What size of a team will we need to hire in order to review all these inspection reports? Our AI through NLP is able to read through all the inspection reports, cross-reference this with all of our approved underwriting guidance by state, and then flag any mismatch to our underwriting review team, saving us hours and hours and hours of human work, as well as eliminating human error.
This looks like a great house. Who wouldn't want to insure this house? When I just joined Lemonade, our chief claims officer told me, "Maya, when it comes to homeowners, it's all about the roof." Looking at this picture, you might miss a dimension which is really important in how we understand the risks for different homes. Through computer vision, our AI is able to look at the rooftop, not only identify any sources of disrepair and flag this to our customer and make sure that they fix that before they become our customer, but also the type of the roof, which might suggest even the type of the structure and other nuances that we're able to learn through that. Let's move to pricing. Pricing is important in two ways. One, obviously it's important for profitability. You want to make sure that you're pricing your customers right.
It's also important for competitiveness. The more granular you become, the more accurate you become for each customer, then that impacts those two elements dramatically. We've been focused in the past 12 months to increase the sophistication of how we price our customers. A lot of that has only been able for us to do in terms of the datasets that we have really in the past year and a half or so. This is pretty fresh. Let me give you some examples of how the transformation we've done in the sophistication of our pricing. When we started Lemonade, and this was still true for about a year and a half ago, we were using territory factors of counties. California has somewhat of 58 counties. Try to imagine the different types of risks that you have in one county, pretty wide.
In our recent California filings, we used 200 square meters, which gives us about a million territory factors, which is 17,000th time the granularity of what we had before. We also care about the catastrophic exposure we get for our customers. Before, we were using wide zip codes and sometimes even street level in terms of catastrophic exposure. Think about a street that is built on a hill. Should the house at the top of the hill receive the same score as the house at the bottom of the hill? We've now moved to address-level catastrophic exposure. Every single home receives the appropriate score, both in terms of pricing and going back to what I said at the start, feeding into our LTV, making sure that we also manage our aggregation in a good way when we're spending marketing dollars. Daniel shared this with you.
We've completely changed the structure of how we file our pricing. Most insurance companies use linear regression, which basically takes two different factors, correlates them together over time for your risk. We've now transformed into a machine learning model, which allows the model to decide and find correlations all on its own and create much more accuracy in how we're pricing our customers. There's some things in terms of the correlations that I can't expose, but they were pretty surprising. We spent hours with the Texas regulator reviewing the machine learning models, making sure that they're both accurate and fair for our customers. To the best of our knowledge, we are the first insurance carriers to use machine learning models in approved pricing for customers end to end.
Going back to the predicted loss ratio, I said we use it also to find correlations between the initiatives and decisions we're making, and we want to see that impact in a predicted loss ratio. When we launch these rates, we would hope to see an improvement in the new cohort predicted loss ratio. It's great that this is really what we saw. We launched these rates in Texas, September 19, and we saw a drop of almost 20 points in the predicted loss ratio for all these new customers buying through these new rates. The best news is our conversion rate slightly increased and the total volume of sales we've received from Texas remained the same. This is really telling you that it's not that we reduced the number of policies we're selling, it's just the granularity and sophistication of the pricing became better.
I just showed you some of the work, not all of it, that we've done in the past 12 months. How much is it reflected in our actual loss ratio? Very little. This is gonna take time. That lagging effect that Daniel spoke about as well is really in full effect here. It's gonna take time until you see it in our actual loss ratio for homeowners. We believe that we're on the right path, we're making the right decisions, we're pricing our customers appropriately, and you will see it will show in the next few quarters. I like to joke and say that the best thing that happened to home is car, but it's pretty true. In the past six years, we were selling renters and homeowners essentially with one hand tied behind our back.
The way homeowners insurance is sold in the U.S. is bundled, and we didn't have that. Launching car was instrumental and very impactful for all of our business. It's a huge market, and from where we stand right now, essentially unlimited, and we're seeing huge pent-up demand from our customers and from people who are not our customers for us to come and launch Lemonade Car in their state. In fact, the number one most asked question for our customer care team for Car is, "When are you coming to my state?" We have over 300,000 customers in our car wait list. Growth is not top of mind for us. We believe that would be something that we'll be able to grow with. We want to make sure that we're growing right and profitably.
In order to do that, we have to make sure we have the right data and that we're able to price and underwrite our customers in a good way. We also don't wanna wait six years like we waited with home. Not surprisingly, and this is something we shared quite a lot about, the acquisition of Metromile was really essential in helping us building these datasets. Billions of miles driven, 10 years of driving data that we received from Metromile that we're still in the process of integrating into our systems. In fact, next month we'll be launching our first LTV model for car. It took us four years to do that for home. Really crunching time and being able to get faster to the area of accuracy for our customers. The second thing that Daniel touched on is telematics.
We have 90% telematics. Across the U.S., there's just 4% adoption of telematics and 2% who keep it on ongoing. At first approximation, every Lemonade customer is driving with telematics and no one in the industry is. This is very powerful. Let me show you why. The way the industry prices customers for car insurance is by proxies. It looks at your age, your marital status, your credit score to try and anticipate what type of driver you're gonna be and how well are you going to drive and what are the chances that you're gonna get into an accident. We don't use proxies. We use how you actually drive in order to determine your driving quality, in order to decide what type of risk you pose, and in order to understand what is the probability of you having an accident. That is dramatically different.
Now I said we're still in the process of digesting a lot of the Metromile data, integrating that into our systems. We also had the car insurance product just one year on the road in just three states, so this is early days. I wanna show you some examples of the things we're already doing with this data. We know when our customers have had an accident. This allows us to initiate the first notice of loss automatically without waiting for the customer to reach out to us. We don't have to build, like other carriers do, huge teams that need to reach out to the customer, chase after them, and try to fill in the gap of data of what happened between when you had the accident until you actually decided to file a claim.
We're also able to be there for our customers when they need us. We prompt a digital prompt for our customers when we know they had an accident. We ask them if they're okay, and if they don't respond, we automatically trigger emergency services to reach out to them and be there for them when it matters. We're able to identify if other people from your household are driving your car as well, people that might have never gone through the quoting process and have never been priced. Let's think about a household with a 47-year-old mom with a perfect driving experience history and a 17-year-old kid getting in a car for the first time. Who do you think got the quote?
It's no surprise that when we interview people through our claims team for car, they actually say this is one of their biggest problems they have to deal with. This creates huge premium leakage, and obviously, you're mispricing your policy. By being able to know that, we're able to better price our customers. We're also able to know and identify if you're using your car for something other than personal use. Rideshare is a good example, and that triggers for us underwriting reviews. Very similar to home pricing is a huge focus for us when it comes to car insurance. I spoke about sophistication for home. Let me speak about the volume and velocity when it comes to car. You want to make sure that every time you learn something new, when you have new insights, new data, that is reflected in your pricing.
Some of the regulatory process of pricing is out of our control. Takes time, different regulators, different states to approve rates, but whenever the ball is in our court, we want it to go as fast as possible. We don't wanna have any lingering. We've developed a tool for our insurance team to be able to, with a click of a button, update the entire rates and flush them through the book. We used to require backend engineers coding. It used to take us months. Now it takes us literally days to update rates. Every time we learn something new, we're able to reflect that in our pricing. This is really building us for scale, as Daniel mentioned, making sure that we have all the tools, platforms, and infrastructure that we need so that the volume of these products grow.
We're able to price them, underwrite them as accurately as possible to make sure that we hit profitability. Car is already, but will become more and more one of the most important levers for us to increase our bundle rate and cross-selling for our customers. I'm glad I've had the opportunity today to share with you in more depth the types of decisions and the sophisticated technologies that we're deploying. I hope that you now share, or at least understand our convictions, that the systems and the processes that we've built internally are sending us on the right trajectory. With every turn of that flywheel, our predictive data continuously grows, our machine continuously learns, and our pricing and underwriting become more refined and precise.
This is putting us on a path to reach parity with some of the more established players in our industry and then go beyond, which will have far-reaching implications for our business. To get a better handle of how this trajectory looks like from a financial lens, let me hand it over to Tim Bixby, our CFO. Before that, here are 60 seconds about a topic which is an immense source of pride for all Lemonade employees, the Lemonade Giveback.
I'll take you through some of the metrics and the numbers and some of the modeling we've put together that underpin much of what we've talked about today, but also much of what we're thinking about in the days and quarters to come. I'll walk you through a five-year model and a couple of variations with some sensitivity that we think will be very interesting and useful, for most of you. We'll talk about cash, capital availability. We'll talk about some loss ratio detail. We talked a bit about our predictive loss ratio, our predicted future loss ratio. We'll talk a little bit about dig into our actual a bit more. We'll talk about our path to profitability. We'll also touch a bit on our surplus strategy, one of the areas of capital use for us.
I hope through the next few minutes, it'll give you a better feel for some of the numbers and metrics underpinning what we're planning and what we've shared today, and why we're confident about the next phases for the business. Now, a thing to think about is the environment we're in. The model that we're gonna share, a path to profitability, getting to a place where the capital that we have on hand today gets us to cash flow positive, to break even, we have to be aware of the market that we're in. The market that we're in has certain conditions. We plan to put ourselves on a track where we are not forced to raise additional capital in a market where that is pretty expensive. Two years ago, the world looked very different.
Today, we have to be working within the reality of the market that's in front of us. Now, this is not a market of our choosing, and someday it will change, but today we're basing our decisions and the model we're sharing with you on that market. Three main levers to think about in the cases and the scenarios that I'll walk through. Growth is a key one. We've modeled out a 20% compound annual growth rate over the course of the five years. Gross loss ratio, a key lever for us. We talked about the predicted loss ratio and the cohorts coming in looking quite positive, seeing that nice improvement quarter-over-quarter. We've targeted over the five years in the models that we're gonna share with you, getting to a 70% loss ratio by year five.
If we think about that multi-line customer rate, what percent of our customers have more than one policy? We showed you 3.7%. A huge opportunity ahead of us in that line. Incumbents, best-in-class performers see a number more like 60% of their customers with multiple policies. In the modeling, I'll show you getting to a point of about 25% over the course of five years. Based on these assumptions, here's a base case view, and I'll walk you through the key parts of this step- by- step, and then we'll look at some sensitivities on this case. We'll end the year with roughly a little less than $1 billion in cash equivalents and investments. Now, some of that is restricted. Some of that is set aside for surplus, as is common for every insurance company.
Think of that as the starting point going into next year. We have noted a number of times over recent quarters that we expected Q3. This past quarter that we just reported to be our quarter of peak losses in terms of the EBITDA loss that we reported. We continue to believe that that is the case. Over the course of the five-year model, this base case, you can see those next five bars is that EBITDA loss declining consistently year-over-year, till in year four you see a pretty small bar and then flipping positive in this base case. What this tells you is five years out, you'll see a cushion, a sufficient cash to get us to that point. Now there's a couple other factors to keep in mind that we factored into this modeling.
There's capital expenditures along the way. That's fairly nominal, but more than zero for us. That's a use of cash. There's also the opportunity for interest income, investment income, and that's a nicer environment today than it was not too long ago. We factored both of those in, and the net of all of that is a bit favorable to cash, and that's the sixth bar that you see there. The last piece that I think is worth noting is working capital. Historically, the change in working capital has been favorable to us. We collect premiums upfront. We pay out claims over time. Pretty straightforward. That tends to give you a source of cash. We've modeled that as zero in this forward-looking case. Over time, that could be an additional source of cushion for us.
Again, that's the base case. Three key drivers. Think of the growth rate. Again, the growth rate is at a pace of our own choosing. We deploy capital to acquire new customers, and we can adjust quickly, and we can manage that in a way that enables us to be thoughtful about how we deploy capital. For this model, in this environment, with the goal of getting to that point with just the capital we have on hand today, moderating with a 20% growth rate. 70% gross loss ratio and that 25% multi-line customer rate. This is that base case. Same case, a little different view, sort of a dashboard view, so we can compare a couple of other cases that we did.
Again, over the course of that five-year period, getting to a point where the trough is about a $100 million minimum level of unrestricted cash. Think of that as sort of the answer to the base case. Then about four years out, mid-2026, turning from a cash EBITDA loss to a profitable position. At the end of year four. Let's look at a case now where things go a little bit better. We get a little bit more of a tailwind in the model that we've put together. Let's keep the growth rate about the same at 20%, but then let's moderate that gross loss ratio. Let's say that we're able to get to a 65% gross loss ratio.
You saw numbers better than that in the predictive numbers we looked at just a few minutes ago. Let's say the multi-line customer rate tracks a little more towards that industry best practice of 60% at 35%. Lots of room yet to grow, but showing some nice upside in that number. What does that do? Well, the $100 million of trough cash almost doubles, $175 million now in this case with these additional tailwinds. Turning profitable almost a year earlier, 2025 versus 2026. Obviously, this is an easier case to manage when these things start to happen. We won't see these at the end. We'll see this as the years go by. In this case, we could obviously consider accelerating growth beyond the 20% with that additional cash cushion.
Now let's think about if things don't go quite as well as the base case. Maybe there's a little bit more of a headwind on some of the metrics. Again, keeping the growth rate consistent at 20%, so we can have an apples-to-apples comparison. Let's say the gross loss ratio is not tracking to where we expect it to be, and it caps out at, say, 78% versus 75%, 70%, 65% in the other cases. Let's say that multi-line customer rate continues to improve. We're on a nice track, but it doesn't quite get as far as the prior case. Say it gets to 20%. The result of this case now, you see what was a $100 million cash cushion becomes a deficit. A $125 million unrestricted cash deficit at the end of the five years.
The good news here is it's relatively small sensitivity. $100 million is not half a billion dollars. It's not a billion dollars. We would unlikely watch these metrics unspool and do nothing about it. We would see this coming over the course of the earlier years, and we'd be able to make choices. Perhaps the market environment is a bit different. We're able to raise more capital. That's an option. Or perhaps we moderate growth. Let's show what that looks like. If we pulled back growth even further, let's say we were in a market where we chose, for the cost of capital reasons, not to raise more capital. We could pull growth back to, say, 12%, and the result is a transition back to that $100 million cash cushion that we saw in the base case.
Again, profitability coming back in to about the same point at 2026. Now, to be sure, this would be a smaller company if the growth rate is dialed back a bit. The big picture, the kind of takeaway I'd like to leave you with from these various cases is that we have ample capital to control our destiny. Because of what we've done in the past couple of years, we moved quickly to raise capital in a market that was conducive. We have the benefit today of not thinking out three or four or five or more quarters, but thinking out three or four or five or more years. Underneath these assumptions, we have some confidence and some metrics, and we'd like to share a little bit of that with you.
Why do we think we can navigate this path? Well, a couple of things, some of which you can see externally. This is a view of our in-force premium over the course of the last five years. Steady, smooth, looks like a straightforward business tracking nicely over the course of the years. Think of the macro trends happening during the course of this period of time. A pandemic. Shifting a company from private to public. Significant inflation we haven't seen for decades. A bear market, a war in Europe. Yet, with these macro headwinds, Lemonade steadily moving upward and improving and growing over time. Some suggest that insurance is recession-proof.
That may be a little bit extreme, but Lemonade in particular within the world of insurance, I think, has shown terrific and impressive resilience during this period, particularly when you think about some of the macro headwinds. We have a fair amount of visibility. We can see what's coming, not forever, but pretty far into the future. In fact, in just the time since we went public about 2.5 years ago, each quarter we come, as most of you know, and we give our view of the coming quarter. All of our key metrics, how we see them spooling out. In 10 out of 10 quarters since going public, we've been able to hit our own guidance or exceed our own guidance every quarter.
Now, this is the kind of thing you don't want to brag about too much, because it's a record that may not last forever. The reason that we have this visibility is just the nature of our business. Visibility is the nature of our business. We are large enough, we are multi-product, and so we can see into the evolving book of business with good confidence. We talked a bit about our predicted loss ratio, the loss ratio of the cohorts that we're acquiring today, and how we expect them to evolve. Our models continue to get better and better. We're also seeing this in the actual loss ratio, the lagging indicator. The forward indicators are super important. That's how we manage and run the business. The lagging indicators are facts, and we have to deal with the facts. Now, the facts are pretty good.
This is our view of our Home, Pet, and Renter's loss ratio. By product over the last four quarters, actual results. Fourth quarter is obviously not actual yet, but we have a feel for how that might play out. Now, what's not here is notable. Life is not here. We don't write those policies on our paper, so we don't manage those claims. Life is not on the chart. Car is relatively new, relatively young. The product we launched a year ago is still relatively small and evolving, so not quite at this mature stage of these products. Metromile, certainly brand new to us. We haven't had much of a chance to impact that yet.
Those numbers are public and available for all of us to look at from the first part of the year. A consistent trend, even in an interesting year from an insurance perspective. Home tracking from a 132% loss ratio heading towards 117% estimated in Q4. Pet, 93%, heading to the mid-80s, 86% by Q4. Renters, the most mature aspect of the book, significant piece of the book, 61%. A little uptick there because of weather conditions in the recent quarter, but tracking to the mid-50s by Q4. Good progress, a consistent theme and continued support for the fact that our predicted loss ratios are something that we can also rely on. I wanna talk a little bit more about path to profit.
Path to profitability is on the tip of everyone's tongue. Everyone wants to see it, everyone wants to feel it, ourselves included. Now, we're not here to just get profitable. We're not here building Lemonade to just break even. We're here to do something much bigger. I think a couple of the charts that Daniel showed showed the potential future vision of what can happen if your business grows and continues to grow at a pace that we have. We wanna show you our path and how we're tracking this environment. The path is really here in front of you. EBITDA is a good proxy for cash flow. It's a number we report every quarter. It takes out the non-cash charges and the significant one-time charges from operating income.
Here you see EBITDA divided by our gross earned premium. The earned number, but it's before the impact of reinsurance. Quota share reinsurance that we have and others have can move some of those other numbers around. This is sort of an apples-to-apples comparison of how our EBITDA investment, our EBITDA loss is improving consistently over the past five years. What you can see here is our, and I think you've heard it a couple of times today, our most significant investments are behind us. Not all of our investments are behind us. Our most significant investments are behind us. The platform we've built, the multiple products that we've now launched just in the last few years, the integration of our first acquisition, significant investments now in the rearview mirror.
Now, let's look at those same numbers from the right-hand side, and let's blow out the right-hand side of that last chart. Same, similar trend. Now you see a little bit more volatility quarter-to-quarter, and that is the nature of our business. That's the nature of insurance. Look at that trend line. Even in this market, going public, launching Pet, launching Car, first ever acquisition, right in the middle of integrating with a whole additional public company, steady upward progress up and to the right. We also invest a fair amount in what I consider our most valuable resource is our employees. The people who build incredible products, the people who create a customer experience that we believe is second to none in the industry, even that investment is becoming more efficient.
For a long time, we were investing ahead of the curve, ahead of growth, because we had to build these things. We had to build a regulatory environment inside the company. We had to build an infrastructure. We launched a business in Europe. We had to expand in states all across the United States. That has now shifted. Just in the past year, you can see what was a pre-investment is now starting to show significant leverage. What is on this chart here is in-force premium, so that very, very top line, the size of the book of business, divided by our number of employees. Over the last four quarters, 25% improvement in that metric alone, just in the past year. Our customers are amazing, yet our average premium per customer is $343.
That's an incredible achievement. Five years ago, a little bit after I joined the company, our average customer paid us $138. Some of our first customers paid us $60 a year. In just five years, 2.5 times greater premium per customer. In the most recent quarter, a bit of an uplift from the Metromile combination is in there, of course. Even before that, consistent upward growth in that number. The glass is. I'm not sure if it's half full, but the glass is going in the right direction on this page. Significant improvement. Now let's back that up and look at the entire market potential. That same chart from the last page is now shrunken down to the left side here, and it's barely visible.
That $343 of Lemonade average premium, it's there, you can see the arrow, but you can barely make it out on the page because this is the market we're in. This is the market potential that we're going after. The U.S. average, maybe a home policy and a car, one car policy, that's a greater than $3,000 premium per customer. That's the average. The top Lemonade customer, this is interesting. Top Lemonade customer today pays more than $10,000. So we're there today. Now, there's only a couple at that level, to be fair, but we're there today. We have the product, we have the portfolio, we have the support mechanism for customers to pay us $10,000. If we dial back just to the point of our IPO, that number was $6,000. Let's have a magic wand.
Let's wave our magic wand. Let's say we added no more customers and we moved everyone to the average. That's a 10x business, a $6 billion business. We wanna talk a little bit about surplus. There's a lot of insurance experts in the audience, and we wanna talk a little bit about how we think about capital management. A reality of the assets that we have is that some of them have to be set aside. They stay in our bank account, but they have to be set aside and restricted for capital surplus. Every insurance company has this regulatory requirement. Important to note that it is still our money. It sits in our bank account. We can earn income on it. It's still on the balance sheet as an asset of the business, but it is restricted.
We can't go out and acquire new customers with it. We've simplified down some basic metrics to help you just get a feel for how we are thinking about it, how we've modeled it, how we factored into the model that we shared today, and how we're thinking about it going forward, because there are some things that are evolving. There's a simple ratio I'll use to kind of take a little bit of the complexity out of surplus requirements. There's a deep, complex model for every company that drives what your actual surplus need is. It's got a lot of moving parts. At a very high level, with no real thoughtful management of capital or surplus, think about a 3-to-1 ratio.
What that means is, for every $3 of premium that we go out and write, $3 of in-force premium, we would have to set aside $1 of surplus into this restricted account. It's ours, but it's restricted. We can't use it. Now, we're gonna be a little more thoughtful as others are, and we're actually in the process of setting up what's called a captive structure. This will be new to Lemonade. It's not too new to other insurance companies. It's a commonly used approach. It's a legal entity, a regulatory structure that enables you to get more leverage out of this capital surplus ratio. It's a little more common at companies that are growing a little bit quicker.
If we employ this captive in the way that we expect we'll be able to, we can expand that leverage from a 3-to-1 ratio to a 5-to-1 ratio. Think about instead of setting aside $1 for 3, we set aside $1 for every $5 of premium that we write. There's also quota share reinsurance, which is a structure we have in place today and we've had in place for some time. Today, we cede or share about 55%, exactly 55% of our book of business with our reinsurance partners. That's been as high as 75% in the past. Reinsurance is a market just like any other. We expect that reinsurance will be available to us. We don't always know what the cost will be, what the price will be, what the terms will be.
If we find them to our liking, that is a lever that we will take advantage of. If it's not, we have other opportunities and other ways that we can manage our capital. Think of the reinsurance lever as something that if we find terms attractive for as much as 75% of our book of business as we have in not too distant past, that ratio becomes 8-to-1. Think of a 3-to-1, a 5-to-1, and an 8-to-1 as potential ratios for this surplus structure. Now, the reality will be it'll be a combination of these, and we don't know exactly what that combination will be. We'll be able to augment and adjust over time as makes sense and as terms dictate.
I would think overall is sort of a mix of those potential structures of about a 6-to-1 ratio. That's what we've factored into the models that we have shared today. Now, underneath the detail, there's more. In the surplus world, there's things like risk-based capital, RBC. For those of you who are experts, that adds a little bit more to the surplus, and we factored that in. There's a little bit more requirement for companies who are fast-growing or for companies who have had historical losses. These are all true of Lemonade, and we factored that into the modeling that we've shared today. Back to one of my first comments. We're not here to break even, although we expect to break even. It's a checkpoint along the way. We're here to build something big. We've already built something great.
We've built a structure and a product and a user experience that we think is second to none on the planet in the world of insurance. We're on track to build something that's much bigger. If you maybe do a little what if experiment with me, I think it would be a little bit fun. Today, we have 1.8 million-ish customers, $343 on average, about a $600+ in-force premium business. Let's imagine we double our customer base. We've done that in just the last 18 months or so. Let's imagine we double our customer base again, and let's imagine that we take that average premium just up to half of the national average, 50% of the national average.
That's another 10x business, 10 times where we are today, $6 billion in-force premium. Let's roll that tape forward a little bit further. Let's say we double it again. We double once and we double again. Let's say we take the average customer with multiple products that we're able to offer up to just 100% equals the average in the US. All of a sudden, that $6 billion is now a $23 billion premium business. Let's apply a 12% operating margin. Let's apply a couple more points of investment income. That looks a little better today than it looked a couple of years ago. I think that can give you a feel for why we're here, what kind of value and potential value we think that we can create.
If we double again, and if we move that average price point up in the way that I described, we'd still barely be breaking into the top 10 of insurance companies in the U.S. We'd have a less than 5% market share. This is why we're here. Break even is nice. Path to profit's even better, but this is why we're here. Next up is Q&A. Before we get there, we have two minutes on something that is very near and dear to the hearts of all Lemonaders, and we hope you as well.
2 billion people in the developing world rely on the crops produced by smallholder farms, crops that are more likely to fail every year because of intensifying climate change. A quarter of the world's population is at risk. In Africa alone, there are over 300 million smallholder farmers becoming increasingly vulnerable. This is exactly what insurance is for. Protecting smallholders isn't financially viable for traditional insurance companies, which is why crop insurance is either too expensive or nonexistent, leaving 97% of farmers in low-income countries without a safety net. Until now. The Lemonade Foundation is proud to introduce the Lemonade Crypto Climate Coalition, a new nonprofit initiative. It's the world's first Web3 organization built from the ground up to protect subsistence farmers from climate risk. Blockchain enables radical new ways to overcome the limitations of traditional insurance, all with low costs and a minimal carbon footprint.
Our decentralized autonomous organization uses highly accurate insurance models, has replaced policies with smart contracts, and uses oracles instead of claims professionals, allowing us to do what was once impossible, accurately quantify climate risk and deliver instant and affordable protection automatically at cost to the people who need it most. Best of all, it's easy to access. Farmers can buy coverage from their phone, paying only their expected loss, and get paid automatically if their crop fails. No need to even make a claim. Here is Eunice Jesang on October fifth, paying just $0.83 to protect her maize crop in Kenya against failure. Traditional insurance needs local teams, but the Crypto Climate Coalition is infinitely scalable, ensuring a more stable future for billions of people. With the possibility to reach rural Kenya the same way it reaches Peru or Papua New Guinea, this is more than just insurance.
It's community on a global scale.
All right. Nice. We'll be taking questions from here in our offices and also online. So if you'd like to ask a question, please raise your hand and one of our team members will come over with a microphone. I'll start us off online as we get warmed up in here. Tim, I'm getting a question from Darren Stark and Pierre Quasney. Both are talking around our G&A spending, asking if it's possible to cut it down by 50% and still support mission-critical operations. In general, what measures are we taking to improve it?
G&A spend, a topic that's near and dear to my heart. I should note that we have talked a fair bit over the recent past about peak losses. Q3 being our period of peak losses. What a period of peak losses means is you continue to invest during that period. G&A is no exception. Q3 was a bit of an anomaly. We merged formally on July 28 with Metromile. In Q3, you've got two things happening of note. One is the transaction itself closed, and in the numbers of Q3, there's $7.4 million of expense related to the transaction, one-time expenses. That's detailed in our disclosure, in our shareholder letter, and for more detail, I would encourage you to look there.
It's also the quarter of, again, bringing the companies together, so you get sort of a double penalty. We've done quite well, I think, in optimizing some of those expenses, in the folks that we've brought on, the synergies we're seeing between the businesses. I would think of Q3 as sort of that maximum point because we're bringing those two companies together. Then over the coming period, we'll see things optimized, and our guidance in Q4 indicates our kinda first next look at how we see the expense lines evolving.
Okay. I'm gonna remind that you can raise your hand and a member of our team will come over. Maya, a question from Thomas Leighton, who's asking what kind of actions we're taking to reduce loss ratio.
I think I shared quite a lot in my presentation in terms of the progress of both sophistication, and a little bit of the granularity that we're now able to do in our pricing. Also each of our product lines needs different attention and different actions that we're taking. Renters is already profitable. A lot of the focus that we're gonna have is around rates, both in terms of velocity, also in terms of the sophistication and granularity. There's a lot of that happening, already and will happen in the next few quarters.
Okay. Tracy?
Tracy Benguigui. Is this on? Tracy Benguigui, Barclays. Tim, you mentioned the possibility of setting up a captive, and I recognize New York, your lead regulator, is a pretty tough one. Have you began conversations with your regulator? I'm just wondering, would the captive be offshore?
It's relatively early in the evolution of the concept, and we're at the point where we wanted to introduce the fact that it's something we have been working on. We're not yet to the point where we're disclosing any of the particulars around how the formal structure will be put in place, and that will, I would expect, evolve over the coming months. We'll, to the extent we can share those details, we will. At this point, we're just stepping into the structuring of it.
Okay. If I could just ask another one.
Yep.
On your predicted cohort lifetime loss ratio, you gave that slide from the progression from the first quarter of 2021 through the third quarter of 2022. I'm just wondering what the retention rate is of those customers. Like, what should we typically expect in terms of cycling out those customers, so you get the fruits of LTV?
Yeah. A couple of thoughts. One, Tracy, just coming back to your earlier point, you're understandably focusing on New York. Do remember that as of last quarter, we have two regulated entities, one of them is in Delaware. We do have more optionality around that as well. We disclose our retention on a dollar basis. In the last quarter, you saw actually a nice spike in that. That is our dollar retention. If memory serves, we're at 84%, which means customers that were acquired during that cohort, their dollars now represent, or dollars that they paid today represent 84% of what they paid us on day one.
Okay. That's not blended by cohort. That's cohort.
That is a 12-month rearview mirror. You're looking at the same people, what they're paying you today. In that sense, it is cohort.
Okay. Thanks.
Yaron?
Thank you. Yaron Kinar with Jefferies. Maybe a couple of questions on the reinsurance side. First, when you talk to the reinsurers and, you know, talk about the terms for the upcoming renewal, do they factor in the predicted loss ratio, or are they looking more at the rearview mirror?
Hi, Yaron. Some do, some don't. The reinsurance market is going through unusual tumultuous times right now, and it's certainly been hardening. I have to say that we have, at this point, pretty deep and meaningful relationships with our reinsurance partners. Our largest reinsurer today is Hannover Re, with whom we do a lot of interesting things. The video that you just saw before was of the Lemonade Crypto Climate Coalition. Hannover Re is our partner in that. We do have deep relationships. Our anticipation is that they have a deep understanding of our business. The stuff that we've shared with you, they understand deeply. Our anticipation is that the partners that have been with us will want to continue to be with us.
Maybe a follow-up to that. Given that we are entering a hard reinsurance market and that potentially means, you know, higher reinsurance costs, I understand you have the leverage to maybe scale back on reinsurance, but how ultimately does that affect your gross premium growth and the kind of path to profitability and to cash breakeven?
Yeah. We haven't entered renegotiations around our reinsurance yet. That contract falls due at the end of June. As you probably know, it's customary to do those things 30, 60 days out at the utmost. It's still premature for me to answer with any specificity the question that you're asking about what would the terms be, and the market has been shifting, so I wanna be cautious. We do approach reinsurance, though, going forward slightly differently to how we did early on. You heard us talk about it as a tool for surplus optimization rather than as risk management. We feel pretty good about the risk management piece of this.
We feel like our business is now stable enough, diversified enough, both geographically and across product lines, and we have a handle on what we need to do in order to manage the loss ratio. We are less interested in paying and margin stacking our business, if you like, in order to shift risk. We're much more interested in doing that as a way of optimizing capital. That's why we'll come down to a calculation nearer midyear 2023 of what are the actual quotes that we're getting. We'll have to give up margin in order to get more capital efficiency, and then it will just be us in an Excel spreadsheet figuring out the optimal mix.
Let's take two more here, and then we'll go back online. Jamie?
James Inglis, PhiloSmith & Company. In your slide, you had a predicted loss ratio in the first quarter of 2021 of 86%. At this point in time, what did it turn out to be? I mean, if you were to look back on those policies now, where would that have been?
It's a great question. Thank you, Jamie. I don't have the precise number of that particular cohort, but of course, the actual results are what train the model itself. So I'll answer you kind of in a more theoretical framework because I don't have the specific number to hand. I'd also remind you that that cohort is still only one year old, and the first year, as I explained, is going to be above its average loss ratio. The average loss ratio generally materializes in year three of our customers. So I wouldn't expect that cohort to have an 86% loss ratio over the last 12 months. I would expect the model would have predicted a higher loss ratio over the last two months, but trending downwards.
Because the model trains itself on the actual historical data, I would imagine that they tracked pretty closely to that, but I can't give you a specific answer to your question.
Over time, has the predicted loss ratio, the ultimate loss ratio become closer to what you predicted, or is it about the same?
No, over time, it's definitely asymptoting in the right direction. Maybe, Jamie, I'll add two more points about the predicted loss ratio. One Maya touched on and one we haven't spoken of. The predicted loss ratio does predict claims and churn and all the things that I spoke about, and it does understand that we are in a high interest rate environment, and therefore it will apply a discount for the cash flow. It doesn't know how to predict inflation. If we had a model that could truly predict inflation, we'd utilize it in all different interesting ways. This assumes nominal values, and it assumes that our rates will take care of inflation, that we'll be able to keep pace abreast of inflationary pressures, which are a big deal in our business.
I just wanna highlight that that is one factor that could impact actual results and how they might vary from modeled results, and that's not included in the model. The other piece that gives us tailwinds rather than headwinds, Maya did touch on, which is the model, while it is providing a leading indicator, it is fed by historical data, right? It crunches historical data in order to generate a prediction. It doesn't know what's gonna happen tomorrow. We do. Maya already showed you how many filings we have that we have not yet implemented. We know what features we're developing that we've not yet deployed. So there's a lot of tailwinds still coming to the model. The model is imprecise in at least two ways.
One is it doesn't know about inflation that could hit us, and the other one is it doesn't know about all the hard work that we're doing, and that could help us. Which is why I urge us to look rather than at the specific number of 86% to look at the trend line, 'cause then you are looking at an apples to apples comparison quarter-over-quarter.
Jamie, can you hand it over to your left to Emmett for a question? Thank you.
Hi. How are you guys doing? Great presentation, by the way. I just had a question about the model for that neighborhood in Houston. That was a great case study, but I was curious, what did it discover about that particular neighborhood that made it so, you know, the pricing off? Was it, like, you know, the sister neighborhood seemed to be optimized, but that neighborhood was, you know, clearly all off. I'm just curious what it discovered about it.
That's a wonderful question, Emmett. I don't know. I was playing around with the map and finding these interesting things myself. The team will know. I apologize. I don't know. I will tell you at a high level that the answer is gonna be pricing. It's obviously mispriced. Maya spoke about moving. She gave California as an example. We're doing this across the nation where the precision, the neighborhood precision has moved up 17,000-fold over the course of the last few years. We're getting much more granular, which is what you saw in that particular neighborhood as well, that we are seeing the stuff. How that is then fed in, you absolutely need to identify the root cause. Typically in that case, it will have come down to pricing. I don't know, Maya, if you wanna add anything to that.
No.
Okay. Thanks, Emmett.
Okay, thanks.
Sorry. May I ask a question?
Absolutely. Go ahead.
Yes. I'm Yu Kwan. I'm a retail investor. I have two questions. First question in terms of the dollar retention there. That reflecting the in-force premium churn rate, right? I'm just wondering, like, for the actual customer churn rate, I think based on payback, and I'm also calculating myself it's like between 30%-40%, so it's relatively high, the churn rate. I'm just curious, like, what's happening here, and over time are you predicting to see the improvement? Yeah. The customers.
Maybe I'll make a couple comments then, Tim, if there's anything I'm missing out, you can help me out. It's not been a number that we've disclosed, but it's not as high as some of the numbers that you're estimating either. Let me give you a little bit of color on that. The reason that people churn is overwhelmingly important in kind of thinking about churn, and we look at that data pretty closely. I'm gonna miss the exact decimal point on this, but I'll make the point broadly 'cause I'm working off of memory here. Every single customer that churns gets asked a question of, "Why are you leaving us?" Sometimes it's, "Hey, we didn't like how you handled our claim," or something like that. That is incredibly remote.
If memory serves, it's a fraction of 1%. It almost doesn't happen. It happens, but that is not what is accounting for churn. What overwhelmingly is being answered there is stuff like, "I'll be back." It's, "Hey, I." It's so much of our book is renters, and people went out to college, and they got a policy. Now they moved back with mom and dad. They no longer need a policy. They moved in with their girlfriend. They're now combining policies. They moved to a state in which we don't yet operate, so they don't have it right now. I think it's helpful to think about churn as coming in two flavors. Throughout the last few years, I've disconnected and reconnected with Netflix a bunch of times 'cause I moved home and moved country and stuff like that.
I probably appeared on some churn chart of Netflix, but I've always kind of come back. The cable company, when I cut my cords with them, that was real churn. They're never gonna see me again. We feel really very good that the customers that are leaving us are alumni. They are people who we see this in the numbers as well. They. When opportunity comes, when they need us again, they come back to us. Although we have structural reasons that would elevate churn today, a lot of them, for example, are the absence of car. A lot of people are telling us, "Hey, I need to bundle 'cause I need to save money, and I don't have Car." They go off to State Farm.
State Farm says, "Hey, I'll save you so much if you bundle," and the dollars just don't make sense. They're going to a place where we don't yet offer the product. There's structural reasons which will resolve themselves as we continue to roll out our product. Although the number, I think your calculations are high but also understandable, if you pierce one level deeper and try to say, well, what are the underlying causes and concerns, we're pretty bullish on those.
Over the time, are you expecting to see the improvement?
Yes. I think you're seeing that in the-
Possible turnaround, right?
-the dollar trend is reflective of that as well.
Yeah. Yeah.
Yes.
The second question is about the predictor model you are showing. One's a lifetime value model, the other is like the lifetime loss ratio model, right? You have different generations. I'm just wondering, like, what is kind of observed value? I think it's also similar question as that gentleman said. What's observed lifetime value that you calculate the customer that you say acquired in 2020, and then two years later, you calculate it again based on their lifetime value. Is that you're seeing kind of the reflecting your improvement of the model?
Yes. The models are getting better and better. 18 months ago was the first time we were able to make a prediction.
Yeah.
The most we have is 18 months, and our models have matured a fair bit since then. The answer is yes and absolutely, and that is the methodology for how the new generations come about, right? They look back on the results, and they fine-tune and get more and more data. Hope that answers that.
We'll take another question coming online from Arvind from Piper Sandler. Can you talk about the impact of macro in your business? Insurance is an essential product, so it seems like revenue should be resilient. As customers become more price sensitive, should you benefit? And what is the impact on pricing with reinsurance in a tough macro environment? You wanna take that, Tim?
Do you wanna take that one? Me? Well, there's a lot in there, so keep me honest.
Yeah.
I didn't take notes in real time, so Tim addressed some of this, right? This has been one of the big reliefs for us. A young company facing the kinds of tumult that the world has faced, seeing so many companies from all different stripes, particularly young ones, suffer quite a beating and fall on hard times. It's just been amazing that when you look at our financials, it's hard to identify when each of these headwinds kind of hit. It's been a fairly consistent up and to the right, no pivots, no changes of strategy. Investors who saw our presentation pre-IPO or for that matter in our seed round would recognize everything I said today as an expansion and a realization of the stuff that we've been saying all along.
A lot of you have been with us for a long time, and I hope you recognize that consistency. We wrote in one of our recent letters, maybe there was some hyperbole in it, but not much, that looking at our internal dashboards, we wouldn't know that all this stuff is happening in the world. We read the papers, we're aware of what's happening, but it's not reflected very much in our financials other than the one thing that we see on the dashboards is inflation. We've had to pick up, and Maya spoke about this, the pace of filings. In a low inflationary environment, that time lag matters less.
In a high inflationary environment, if you're not filing at a rapid rate, then you can fall into the trap of pricing based on a certain nominal value and then paying claims based on an inflated value. That can be pretty painful. Part of our elevated loss ratio is explained by that. So is, like, our newfound accelerated rate that Maya spoke about and all the energy that was put into the technology to enable us to move from a two-month code intensive implementation of filings to very rapid one. We shift the focus of our energies towards resolving the most pressing problems. Inflation is a real issue. It's not gone, but it is the one that we recognize and contend with.
The second one, which you wouldn't see in our dashboards, but you know this as well as anybody, is just the elevated cost of capital. When our share price was much higher, raising capital was much easier and cheaper. We always wanna make sure that if we are in any way diluting ourselves and raising capital, we wanna make sure that we can give a positive ROI, that the earnings per share will increase as a result of any subsequent fundraising. At this climate, we're not sure we can do that, which is why we're making do with the money that we have. If that climate changes, that will change. That is the other way in which headwinds, macroeconomic headwinds, have impacted our business.
I'd add one to that, which is actually somewhat favorable, which has been our marketing efficiency.
Mm-hmm.
The environment over the past year to two years has been a bit of a step change in the cost to acquire new business for customers in general or for companies in general who are doing primarily digital acquisition, online acquisition of customers. Yet our value, our long lifetime value of those customers has increased faster still.
Yeah.
Even though the absolute cost of acquiring that new business has risen in that period, our net impact has improved somewhat. That's one macro trend that has impacted us, yet we've been able to overcompensate by more efficiency internally.
All right. I'll turn it back here to the floor. A few people back there and Brenna.
Hi. Could you give us a little more specifics about the rate, rates that you're actually going after by product? You know, you talked about the backlog of rates that you have filed for. You know, what are they? You know, what percent increase in rates have you filed for by product? How much of this has been implemented so we kinda know what's in the hopper? In conjunction with that, I mean, all that is to just keep up with loss costs. Where do you think the loss costs are really running? You know, in an auto book or a homeowner's book, they're definitely double digits now. You almost need double digits just to stay even, not even improve.
You wanna start, and then I'll.
Sure.
Yeah.
The car, Tim touched on this in passing, but our own car business only launched one year ago. One year of data, barely seeing kind of annual renewals. It is just very hard to get a good picture. We are seeing declining loss ratios, but the quantum of data is such that we're being very cautious in reading too much into it. It is still very small book, only live in one state for one year and in other places more recently. It really doesn't rise to the level of statistical significance, and therefore, I just wanna be cautious. We are filing, I think Maya mentioned, in Illinois, where we've been for one year, we've already rated three times.
We're monitoring the data, quickly fixing short cycles, rapid iterations, but we're not in a position to really zoom out and give you a sense of what is gonna be needed over the long term. In most of the states, we are still doing our initial filings. In Metromile, the picture is clearer, but not so clear to us yet. In other words, the numbers have been published for some time. They've been at a loss ratio that has been in the triple digits. 65% of that business is in California, and as you well know, they've just not been able to take rate in California. This is an important point to perhaps underline, which is we have absolute clarity on what they need in terms of rate. The data is very revealing in terms of what is needed.
It's just a regulator that is unwilling to allow for that. There's no disparity between what is needed and what we know. There's a disparity between what we know and what we're able to implement, and we respond to that by throttling back on growth. We have spent no dollars promoting the Metromile business since the acquisition. We said that on the day that we acquired them. We will change that the day that the rates are approved. This is exactly what the portfolio management that Maya spoke about before, the fact that there are multiple geographies, multiple lines allows us to do. In home, maybe, Maya, you can-
Yeah.
-to home.
I'll just say, obviously this differs by product and by state. Some of our states require very little rate change, really just to keep up with inflation. The business was profitable even before we went out of all of these rate changes. Just to give you a little bit of understanding of the velocity, we grew our filings from last year for home three times. We're doing three times the filings that we've done last year in our homeowner business. Some of the states that we're seeing bigger impact in terms of inflationary up to 20%, 30% in terms of the cost of rebuilding and changing in the RCE, then we're taking as much as 35%, 40% rates in some of these states. California is obviously a good example, we have a 35% rate standing.
If I can just follow up. You talked. You gave the example of Texas and using the AI models getting approved there. What is the start to finish timeframe that takes to educate a regulator?
Mm-hmm.
In that kind of a model? Yeah.
I think, again, depends on the regulator, depends on the state. I think in many ways, this is a huge focus of ours to really be at the front of it, sit down with regulators, and make sure that we're taking the time to make sure the way we're thinking about implementing these models is in a framework that they feel comfortable with. With some states like Illinois and Texas, I would say we've managed to make huge strides with on implementing that, and now we also have data to make them feel even more comfortable and almost use that as a case study as we move into other states, and get more of that approved in other places.
Yeah. Sorry. Just to add maybe one dimension to what Maya already said. It's an ongoing discussion with the regulators, but it's been ongoing for a while. We do feel that if we want to be leading the industry in terms of AI, it is a burden that we have to lift to educate the regulators and work on this. We do a lot in this. I've met with many, many regulators, spoken to them many, many times. We were joined quite a while ago now by Tulsee Doshi, who is our AI ethicist, who really is meeting with regulators with incredible frequency to try and help them understand the issues and concerns that they have. It's a bit of a gulf that you need to fill because regulators are not trained in AI and therefore their caution is understandable.
We are putting a lot of resources into overcoming that and explaining. We have a very strong thesis that I've written about and spoken about as well, that ultimately properly applied, this is a massive boon to fairness. As you break monolithic groups up, as you stop using proxies, what you're getting is to a much fairer result as opposed to the state of the industry today. It's not that there's a fundamental difficulty here, there's just a lack of familiarity, and we're trying to overcome that. We've also been joined recently, he's in the room with us today, by Scott Fischer, who's in charge of our government relations and was the former head regulator in the state of New York and is putting a lot of his energies into exactly what you're talking about. Thank you.
Yeah. Josh Shanker with Bank of America. As far as I can tell, your customers are having outstanding experiences and, they're very pleased to stay with Lemonade. Their habits are changing and evolving. To what extent are you finding that it may be the case that customers are like, "I'd love to stay with Lemonade," and they go to look at your product and they say, "And not only that, the price is the best." Then your loss ratio is quite high at, you know, 50%, higher than competitors. To what extent, if you got your pricing in line where you were delivering 60% loss ratios in home and pet, would that same customer say, "Look, I'd love to stay with Lemonade, but Lemonade doesn't have the best price anymore," and they're making an economic decision that they haven't had made so far?
I love the question. Thank you. Two or three thoughts on that. One is before inflation kicked in, we reported a combined company-wide loss ratio of 69% just a couple of years ago. We've been there. In certain pockets, in certain states, in certain products, we're there today as well. We can say with some confidence our customers love us and buy at better loss ratios, not just at higher loss ratios. I think the empirical data speaks for itself. Let me expand on that though. What I tried to do in my presentation was take you through the expense ratio and the loss ratio and try to make the case that we should ultimately be able to best the industry at pretty much every line on the P&L, every part of the combined ratio.
The point to highlight is that the ratio part means it is affected by price. What I mean by that is we could raise prices and have a fabulous loss ratio and lose our competitive advantage. The art that we're trying to apply here is how to be a price leader in most places for the right risks that you're trying to get, identify those risks better than others, be a pricing leader for them and not for others. The telematics example that Maya touched on, I think is a fabulous illustration of that point. When you're pricing based on proxies, much of humanity looks very much alike.
You say, okay, these are men and these are women, and men, this is a fact, have twice as many fatal accidents per mile driven than women, so I'm gonna use gender as a pricing factor. But when you use telematics in the way that Maya expanded or expounded, you start seeing that I can see so much more nuance because while men as a category may be worse drivers than women, there's more variation within those groups than between those groups. Therefore it's actually a pretty crude measure, but it's the one that the industry uses.
Suddenly, by getting continuous data streams from the vehicle, I can actually price people based on them being a human being and how they drive rather than them being a husband or a man or a woman or having this kind of job or this kind of credit score, which touches back on my fairness point earlier, but it touches on your point as well. Because what happens then is you take this big cohort of people, and you can see instead of this big chunk of monolithic people, you start seeing all the shades of gray in between. Suddenly the better drivers, the better risks, get a better price than they do than when they go to the competitors' pricing based on proxies, and they will find us very attractive.
The worst risks would prefer to pay average rates next door than pay their true rates with us. You'll start getting a positive selection happening. Loss ratios remain very, very healthy. Prices remain very competitive for the risks that we want and unattractive for the risks that nobody really wants, but that our competitors are unable to identify with that level of precision.
I might add just, I think Texas is also a great example that I gave the new rates. The total impact of that created a rate increase on average. If you look at the impact that we had in terms of the rates that we took, and so we increased prices, but we managed to lower the predicted loss ratio for these groups as well as keep conversion the same and the amount and the volume of business that we were selling. You should expect to see that more state after state as we become more sophisticated. It doesn't mean that we're not raising rates, we're just raising them in a way that is a much more sophisticated and customized for each of the customers, depending on the risk that they have.
We have a question from Katie. I know, Marco, if you could. Alex, great.
Hi. Thank you. Katie Sakys, Autonomous Research. I think this is a question for Tim on the base cases that you outlined. Annualizing this quarter's dollar increase in organic IFP ex- Metromile implies something like a 25% growth rate next year, which seems aggressive to me given that messaging on marketing spend will be decelerating and that 3Q is a seasonally high growth quarter. I'm wondering what's giving you the confidence in today's 20% CAGR outlook, and what marketing spend assumptions are you including in your model?
We of course have not yet given guidance for next year. Today was a bit of a preview, and we have indicated the growth rates of 20% in the model and 20%-25%, that is correct. I think the connection between the amount of capital that we spend, the growth budget that we allocate and the amount of business that comes in is pretty direct. It changes from quarter to quarter and channel to channel, but it's a fairly direct connection and fairly predictable. Now, we're having to deploy fewer growth dollars to grow because we now have multiple products, and we're seeing more and more customers cross-selling, upselling without the additional customer acquisition cost.
In the modeling that we put together, we've assumed that our efficiency is about what it is now, which is higher in Q3, this past Q3, and going into Q4 than it was at the beginning of the year. As I noted a few minutes ago, we're seeing an improvement in that efficiency. We've also seen a bit of a benefit in the acquired Metromile book of business, where our assumptions as to the level of retention that we would have were actually a little bit better in the third quarter than we anticipated, and that's giving us a little bit more of a benefit in terms of the anticipated churn there. I think the combination of those and the sensitivities we've done give us that confidence going into next year.
Now, in Q1 we'll come with more formal guidance as for the quarter and the full year as we have. Between that sort of thinking and the sensitivities we built, really that gives us confidence in that five-year view that even if year one and year two look a little different in real life, that over the course of the model the results are pretty reliable.
Awesome. As a follow-up, I just wanted to ask about some of the variables that might impact those assumptions. What's driving your confidence in the multi-line outlook given that California homeowners is no longer a cross-sell option? Can you point to any specific bundle types that are propelling that growth rate and might be, you know, the stickiest in terms of your customers?
Yeah. Again, it's a tricky one to predict, although we do have a track record now. We can sort of draw a line through two points. We know what that number was a year ago and two years ago. One of the items, one of the slides we presented today showed that uptick for the state, for Illinois, where we have the state with all of the products available. And that gives you that sort of hard indicator that where the products are available and we're able to optimize and enough time has passed for the models to start to learn, we saw that nice steady uptick. Now, again, can we predict with precision, you know, what it'll be each quarter along the way?
That's a little trickier, but I think the benchmark of best in class as 60% is something that's attainable. We're assuming we're less than half of that over the modeling period. Net Promoter Score, customer experience, all of those indicators suggest that we should be able to track towards those best practice levels.
Great. We have time for two more questions. I'll take them from the floor. Yaron?
Thanks. Maybe a couple of questions on bundling. I think if we look at the industry, we haven't seen a lot of success in auto and home bundling through the direct channel. I think that's gonna be an imperative part of the growth opportunity at Lemonade. Maybe you can talk a little bit about why you believe that you can succeed there.
Time will tell. We don't wanna kind of boast of any successes that we haven't yet had. You'll note that in Tim's modeling, he showed quite a range of potential outcomes. We're not giving guidance to a particular number. We're taking a stab in the dark about what sounds reasonable to us over a five-year period. You'll note that we didn't in any way, even on our most optimistic things, we didn't suggest that we're gonna get to parity with just the incumbents. We speak about all our advantages, but the most aggressive model we showed, we'll get to 35% when they get to 60%, and we showed more conservative models still. Our optimism, such as it is drawn from the last 12 months.
Really the experience that we've had in Illinois and now repeated over two other states where we see the trajectory change angle in exactly the same way. It's repeated three times over. That is about car bundling. All that lift is coming from car. The early indications are positive. The trend lines are in exactly the right direction. You can straight line out from that and apply whatever discounts you want to that. We know that there's something of a ceiling in the industry at 60%, and we've kept a very wide berth from that out of conservatism. Let's talk again in the coming years, and we'll tell you how we're doing.
Maybe just to level set the 60% industry average that you were talking about in terms of multi-lines and the premiums per customer. Are those for personal lines carriers only?
Correct.
Okay.
Including direct-to-consumer personal lines, yes.
We have another question in the back. Sorry, I can't see you.
Sure. Two questions actually. One for Dan, one for Tim. Dan, just you made a pretty strong case of AI being the future of insurance, and clearly it's the brains of Lemonade right now. Can you give us a sense of how much you're investing in AI on an annualized basis and growing it? The second part to that, I think to your seesaw where you were suggesting it's front-loaded as an expense, where are we right now in terms of the AI specific expenses on that seesaw? Do you think we are kind of becoming more neutral, or it's still a lot of front load expense in next few years that we should expect?
Tim, I can take a bash at it, but you might be better placed.
Well, I think the way to think about it is it's-
About tech spend in general.
-it's essentially being built by us. Everything that we're investing in, whether it's AI specifically or tech and engineering more generally, of which a significant proportion I would think of as AI or intelligence, is significant. It's about 25% of our overall headcount of the business. Tech, the cost of a tech expert are obviously higher than average cost, so the dollar amount is higher still, in terms of the scope of investment in the 30+% range. I don't have a hard number of what subset of that is AI focused, but I would think of it as a significant focus of that team.
We're at the point now where we're not launching significant new products, so most of the efforts of those teams previously was focused on the new product, building and launching pet, building and launching car, now building and launching a pay-per-mile product to match Metromile product. There's not large products like that coming. There may be additional future products coming, but that significant investment is behind us, and that really opens up the potential to expand that team's focus on things that are working, particularly in AI and machine learning.
Maybe I'll just add one kind of vantage point. You saw the visualization of our system that I showed earlier. We can break out how many data scientists we have and what we're paying them. The deeper thing that I wanted to get across is that actually everything that we build in technology is part of that same digital substrate that is feeding and is fed by that machine. This isn't just about how many data scientists. We've got a lot of world-class data scientists, of course. All the data engineering that we do and all the applications that we build out and all of the product flows that we do are being fed by and feed into that same thing. It's really the totality of our engineering and product expense that is driving the cycle that I was talking about.
I wouldn't wanna isolate it just to data scientists, although we've got many and fabulous data scientists as well.
Great. Thank you. We have a lot more questions, coming in online, and we'll have to tackle them at another day. Daniel, some closing remarks?
Thank you.
Sure.
It's been wonderful to spend these past few hours with you all. We hope that we've convinced you of some of the things that we believe in. At a bare minimum, I hope we have convinced you that we believe in them. We believe that we have built Lemonade on an unrivaled technology platform. We believe that if you know where to look and if we share the data, you can see a lot of the impact of that today, but that it will be hard to miss in just a few years' time. That trajectory is one that we firmly believe in. We believe that we are winning the battle for tomorrow's consumers, that our market share among first-time buyers of insurance is strategic for the reasons that I touched on earlier. It is highly predictive.
Perhaps the single most predictive thing of ultimate market share is market share among first-time buyers of insurance, if you're a long-term investor. Nothing is as profitable as growing with those customers. We believe that the combination of those and the structures that we've built around those are highly differentiated, highly distinct, and highly defensible. They are not akin to anything that any other insurance company is doing, and we've been architected very much to support them. Therefore, we believe that as we roll this movie forward, we will be generating an increasingly valuable business, an increasingly large business, to the benefit of Lemonade, its customers, and its investors, those who join us on this journey over time. I wanna kind of temper that by saying we know we're not there yet.
We know that we are up against massively entrenched and formidable competitors whom we have nothing but respect for. Our belief is not at odds with that statement, nor is it mere hubris or mere wishful thinking. I'd like to try and explain why our optimism, what it's grounded in. Lemonade is sometimes referred to as a disruptor. It's not a moniker we often use about ourselves. We have mixed feelings about it. The reason that we have mixed feelings about it is because, not because we have any issue with disruption, but the way it's conjugated suggests that we are causing the disruption. We're not. The disruption that is coming to the insurance industry is coming because of seismic changes in the way humanity is organizing itself.
One of the consequences of that is that insurance companies are losing supremacy over their most important factors of production, statistics and data. It's as simple as that. One thought experiment to kind of drive the point home. If you stopped Joe Public in the street in the year 1700 and you said to him, "Who are the bastions of the world's data, and who is home to its finest statisticians?" He might have said an insurance company if he understood those words. In the year 1800, he would definitely say an insurance company. In the year 1900, no question, his top 10 would all be insurance companies. Quite possibly in the year 2000, if you stopped somebody in the street and you ask them that exact question, "Who has the world's data?
Who has the world's best statistics?" They might have still named an insurance company. Today, toward the end of 2022, not a single insurance company would make the consideration set. Their question would be entirely answered with a catalog of Silicon Valley-style companies. Google might top that list. State Farm won't appear on it. That is the disruption that is coming to the insurance industry. It has lost dominion over statistics and data, and it's in the business of monetizing statistics and data. Lemonade is not causing that disruption. We can't take credit for it, but we can absolutely take advantage of it. That seismic shift, that secular trend is what's creating the opportunity into which Lemonade is entering. Because companies like Google that are built on an engineering culture are eating the lunch of insurance companies when it comes to data and statistics.
Guess what? That's how Lemonade is built. Now, this isn't a quick turn. You ask us, will this reflect in Q1 results? That's not how we think. We're in this for the long term. We want to build something of sustained value for decades to come. I mentioned earlier that some insurance companies that were at this or a similar intersection 300 years ago are now doing $100 billion a year 300 years later. The prize is huge and worth fighting for, and that is really the perspective that we take as we think about what we're building here. That is the root cause of the optimism that we exude, and I hope we do. I wanna thank our investors, present and past and potential. I wanna do it in a roundabout way.
I've been fortunate over the course of our time at Lemonade to meet a lot of the CEOs and leaders of a lot of great insurance companies around the world. To a person, they are smart and focused and analytical, and the analysis that I just shared with you about the seismic changes and the disadvantages and the structural advantages, they know that stuff cold. They tell me that stuff. You get them in an honest, unguarded moment, they'll tell you that stuff as well. It's what keeps them up nights. They share with me the challenges that they have, because having seen the writing on the wall, that would be one thing if their business was doing terribly. You can reinvent a business when it's doing terribly. But they're doing great.
They just see the impending stuff, and they wanna just ride out their tenure before that all hits the wall. They see the impending stuff, and most of them wanna do better than just ride out their tenure. They are thoughtful about what should we be doing about it, and they struggle. This is the stuff of books. Innovator's Dilemma was written all about this stuff. I remember one particular conversation with the CEO of one of the largest insurance companies in the world, and he starts cataloging all the issues that he faces. He says, "I've got an amazing team of executives who are groomed for legacy preservation, not for business transformation. I have a culture that is risk-averse, even though I'm in the business of risk management. Don't think I don't invest in technology.
I invest billions in technology a year, but I don't produce a black box. It all goes into a black hole because it's spaghetti code dating back to the 1980s, written by people who have died and code that nobody knows how to maintain. The list goes on and on. He talks about distribution. I wanna go direct to consumers. Apps are amazing, but am I gonna give up all my business and my 40,000 brokers in order to make that transformation? Et cetera, et cetera. Listening to him go through this litany of woes, I was reminded of a joke.
It's a Scottish joke, and it's about a guy who finds himself in some remote village in Scotland, and he goes into the local pub, and he stops the first bloke he meets and he says, "Tell me, how would you get to Aberdeen from here?" And the guy looks at him, he says, "If I was going to Aberdeen, I wouldn't start from here." The biggest news to me when he was cataloging the mismatches between what he needs and what he has was his investors. This one I hadn't seen coming. He said to me, "I have an investor base who want no volatility and a 4% dividend every year. And if I did the bold things that I know I need to do, they would oust me 'cause I don't have the right investor base." That was striking to me.
The importance of having an investor base who understand and agree with you about what it is that you're trying to do. Which is why for all of us, and for Shai and I, in particular, since day one, being aligned with our investors has been of paramount importance. We don't want anybody buying Lemonade shares who thinks they're gonna get something different to what we think we're building. We want to open the proverbial kimono. We want you to understand what's working and what isn't, and how long things will take. Lemonade is not everybody's cup of tea. It isn't. If you like what you hear, and this all makes sense to you, then hopefully you'll be drawn to Lemonade.
If not, sincerely, we hope you seek your fortunes elsewhere, because we plan to keep doing what we're doing, and we want fellow travelers who see the world in the same way. In that vein, I wanna thank you all for your consideration and for your time, and for those of you who are fellow travelers, for your continued support and encouragement. I wanna just end the day by thanking our spectacular team of Lemonade makers. These are the most outstanding professionals we've ever had the privilege of working with. Bold, customer-obsessed, detail-driven, big-hearted group of people. Building Lemonade over these last few years has, for Shai and I, been the thrill of a lifetime. Thank you all so much. This wraps our day. Thanks, everybody. Have a great one.