Good morning, and welcome to Lemonade's third Investor Day. My name is Mira Chatterjee, and I am a member of Lemonade's Investor Relations team. We are delighted to have you with us today. Before we begin, a few housekeeping announcements. Today's presentation will include forward-looking statements. This includes statements about our technology roadmap, product plans, financial targets, long-term aspirations and illustrative scenarios, and expected timing. These statements are based on our current expectations and are subject to risks and uncertainties that could cause actual results to differ materially. Please refer to the cautionary statement on this slide and to the risk factors described in our most recent Form 10-K and our other SEC filings, all of which are available on our Investor Relations website at lemonade.com/investor. These forward-looking statements are current only as of today, and we undertake no obligation to update them.
We will also be discussing certain non-GAAP financial measures. Reconciliations to the most directly comparable GAAP measures are included in our most recent letter to shareholders, available on our website. In our demo, we will be sharing screens with financial metrics that are illustrative solely of how our product works. They are not intended to reconcile against reported GAAP or non-GAAP measures, do not constitute guidance or a forecast, and should not be relied upon for this purpose. This event is being webcast, and a replay will be available on our Investor Relations site. With that, please welcome Lemonade's Co-Founder and CEO, Daniel Schreiber.
Good morning, and welcome to Lemonade's third Investor Day. I would like to share with you some organizing thoughts for this morning, and I am going to do it in three acts. The first, I would like to talk about some important applications of AI to consumers and the strategic implications of those for Lemonade and insurance more broadly. Then I want to pull back the curtain a bit and show you some of the latest foundational work that we have done in AI and the strategic importance of that to Lemonade. In the third, I will preview with you some of our financial expectations for the years ahead. Let us dive in. In recent months, personal agents have exploded onto the scene. In no time at all, they have graduated from being a curiosity for geeks to a trusted personal aide to the masses.
Each day, millions of Americans use Muse or its ilk, and they delegate to them some of their most delicate legal and financial and personal tasks. With each passing day, their ranks are swelled by about a quarter of a million new users of these technologies. Insurance will never be the same again. I would like to take the next few minutes to tell you why and explain to you how the emergence of personal agents are shifting the strategic landscape decisively in our favor, how the world is coming to us. Start with the fact that this mode of interaction is nothing new to Lemonade. We have been doing it since our founding. When we first sold our policy, it was by AI Maya, a bot, agentically selling a policy in September of 2016.
AI Jim set a world record when he paid a claim in three seconds in December of 2016. During the course of the morning, you will see what a decade of investing in agentic insurance looks like. This image first appeared in our prospectus when we filed for IPO initially in 2019, and what you see in it is our basic architecture, AI agents in orbit around a central brain. Today, you will see this schematic come to life. You will see how our army of agents enjoy superhuman, up-to-the-minute context, empowering them to do everything in our business. All that is to come. I would like to talk first and foremost about your agents. Many of you will already have dabbled with one or other of these personal agents, and if you haven't, you no doubt soon will.
The more intrepid among you may have tried to set up a Clawbot a few months ago, the OG of personal agents, not yet one year old. Together, these personal agents are mounting a revolution in insurance. It is changing how consumers buy and interact with insurance companies, and Lemonade is entirely aligned with the insurgency. We were built for this moment. Everything in our system is MCP-enabled, meaning agents can operate all the building blocks of our system autonomously. Great news for us need not be great news for everybody. You might think any insurance company can sell a policy to a personal agent, even unwittingly, and by and large, you would be right, although it is stunning how many insurance companies still require human-to-human contact to seal the deal. Buying a policy is only the beginning. Personal agents want to do so much more.
They want to customize coverage, change defaults. They want to ask smart questions, submit claims, appeal decisions. The tweet that you see on the stage is from one of our customers 10 months ago. His name is Nikita, and Nikita had a claim for pet insurance declined by us. He tweeted that while he slept, and unbeknownst to him, his Clawbot went and filed an appeal with Lemonade. It was a long and legalistic and well-reasoned memo that we received, and his appeal was carefully considered, and a well-reasoned upholding of the initial decision settled the matter to Nikita's satisfaction, or perhaps to the satisfaction of his Clawbot. Nikita was an early adopter 10 months ago that was an outlier behavior, but is moving very quickly to become the poster child of personal agents.
On its launch website just a couple of months ago when Meta launched Muse, they chose to spotlight an agent filing an insurance claim as the exemplar of the transformative power of personal agents. This is about to become a huge problem for insurance companies. Every customer will shortly have a tireless advocate initiating, scrutinizing, appealing, probing, and threatening their insurer. The U.K.'s Financial Ombudsman estimates that already now, there is about a tenfold increase in complaint and other letters to financial institutions, but the real tsunami has yet to arrive. Insurers hoping to address personal agents and the flood that they are going to generate using traditional methods are bringing a knife to a gunfight. It is different for us. The appeal that Nikita's Clawbot filed was evaluated and correctly and politely and professionally declined by our agents while we slept.
But the ramifications of personal agents go far deeper than simply overrunning insurance companies with a deluge. Deeper even than being unable to sell and service insurance once again the way customers want to be sold and serviced. Personal agents are oblivious to the most strategic moat the incumbency has built, and that has profound implications for our industry and for our company. Our strategy has always been to use technology to drive precision up and drive costs down, to price risk better, service customers faster, and pass the savings onto consumer, ultimately winning on price. You can see our strategy in the schematic on the screen. That was from our last Investor Day, and it's working really well for us. Yet personal agents shift the world significantly, dramatically in this direction. To understand why, take a look at a market where much of that shift has already occurred.
The U.K. offers a useful precursor for the agentic age. Most policies, the overwhelming majority of policies in the U.K. are sold via price comparison websites, PCWs. For the insurer, it's a largely headless experience. No app, somebody else's website, your brand barely seen. Insurers in the U.K. have to win or lose on the merits of their offering. Fancy that. Price, coverage, user reviews, that's it. It is the bare-knuckle cage fight of insurance, and Lemonade thrives in that environment. We have two products in the U.K. at the moment, and on price comparison sites in the U.K., they list position first more often than any other insurance company. Our U.K. business has been going gangbusters. In the last two years, it has grown almost 10 times faster than the rest of our business, itself no slouch.
That gives you a glimpse of what happens when the way insurance is bought aligns with the way we compete. To date, true price comparison sites in the sense of a destination where you can get multiple bindable quotes have been thwarted by the incumbency in the U.S. It just doesn't exist. Google Compare was meant to do this 10 years ago. It was meant to become the Kayak of insurance, but the largest insurance companies blocked them. Aspiring startups, of which there were several, were stymied by cease and desist letters. Incumbents really don't want consumers to be able to compare quotes based on price or experience. They really don't. Why? There are more than 500 personal lines P&C insurance companies in the United States. Incumbents don't want to have to outprice all of them. They prefer to outspend them.
That's a strategy that has been working for them. The top 1%, the top five of those 500 sell about 70% of the policies that are sold. Top 1% commanding 70% of the market. How do they pull that off? That's a stunning thing to be able to say. Since I launched 10 years ago, those five, the incumbency, has spent about $100 billion making sure that you can't drive five miles, you can't watch five minutes of TV without seeing one of their mascots. It's a motley crew. You've got your Cockney-accented gecko. You've got your perky brunette with her apron. You've got an ordinary-looking guy who happens to be friends with NBA stars and NFL stars. You've got Doug, whose partner is an emu. Doug and his emu alone appear on TV in the U.S. about 36,000 times a week.
We laugh, the characters are playful, they bring a smile to our face, but this is no laughing matter. This is big business. They are the pillars on which the whole edifice rests. There is real science behind the incumbent's capital allocation in this regard. When we hear it becomes obvious. When humans look, like us, look for insurance, our limited time, our limited knowledge force us to rely on a cascade of heuristics in order to artificially limit our search criteria. We start with the availability heuristic. We only consider the names that we've most recently seen. You've got the halo effect, which means that we'll conflate a polished advertising campaign with a superior product. Costly signaling means that if you feature an NBA star, we'll assume that you are dependable.
When a brand makes us laugh, we are inclined to trust them a little bit more and question it a little bit less, and that's the attribute substitution effect. So we narrow the field before the comparison even begins, and that is why insurers pour billions into relentless campaigns of memorable mascots. They are buying a place on our shortlist. This has been their most enduring moat, and it's been the hardest one for us to breach. Look at it from our perspective. There's a lizard and a duck and an emu that are better known for insurance than Lemonade. Their strategy, their capital allocation assumes, though, that the buyer is susceptible to these mind games. That was a safe assumption until five minutes ago. But those billions invested in hacking the human brain leave personal agents unmoved.
It's not that an oversized gecko loses his charm, it's that the thing doing the buying cannot be charmed. What do personal agents care about? They care about price, they care about reviews, and they are relentless. They'd have no need to stop after the first three quotes because they've got to go and do something else. That transition neuters the moat, the castle, the whole edifice that the incumbency shelters behind. Friends, it does more than level the playing field. It changes the game, and it changes it to a sport in which we are best placed to win. Astute commentators have already picked up on it.
You've got the CEO of Perplexity, Aravind, saying that, "For objective choices like finding the best insurance, agents will increasingly make decisions on our behalf, reducing the role of ads." You've got the industry rag, Digital Insurance, saying that, "AI agents don't shop like normal policyholders do. Brand image means far less to them than rates and other terms." Chamath put it best, "Brands," he said, "are going to zero." You can see why this is such a strategic shift. There is no other commodity in the U.S. where the brands behind it pour so much into their own brand association. Progressive Insurance alone, last year alone, spent $5.1 billion making sure you all know who Flo is. That is more than the entire U.S. oil, gas, and electricity industries combined. This is big business.
It's early days, but the logic that I'm laying out here is sound, and the reality so far is in line with it. This is the launch page of Hark, the most recent personal agent to launch. Once again, insurance spotlighted as a primary use case. The agent is tasked with finding better home insurance than the current State Farm that the person giving the command has. The results come back. Lemonade beat out State Farm, Progressive, and Liberty Mutual. This is already a big deal, but it's about to get a whole lot bigger. In the couple of years before personal agents launched, already about 1/3 of insurance purchases were informed by AI interactions, people chatting to Claude or to OpenAI, ChatGPT.
But if McKinsey is right, this is going from personal agent transactions are going from almost nothing to $1 trillion by the end of the decade. This is going to be a rapid shift. The world is coming to us. I want to shift gears for a minute. I want to take you backstage, pull back the curtain, and show you how your personal agents fit in within the broader agentic habitat that is Lemonade. Categorically, Lemonade is not adopting AI or deploying AI. It is a network of AIs that we call the Lemonade OS. It's a honeycomb of intelligence. In the presentations that follow mine, you will see analyst agents and you'll see claims agents, insurance agents, product agents. You'll see all of them in action. But first, I want to take you on a tour of some of the foundational layers of the Lemonade OS.
These are what make everything downstream possible. This is the Lemonade OS. We're going to switch to a quick live demo. I'm just going to log in. And what will start exploding onto the screen here is the central nervous system of Lemonade OS, that central orb that you saw in the schematic. Except that this is live, this is real, this is untouched, this is undoctored, except that we've scrubbed the labels in order to protect trade secrets and privacy. What you're seeing behind me is a 3D visualization of the context of all of Lemonade's business. Every speck of stardust here is a piece of information, and its vector, where it sits in the space, tells you something about its relationship to other parts of the business, and you see these synapses where the connections are tight.
I can interact with it, I can pan and I can zoom, and I can click on different modules here. And when the labels are in place, it's a fun thing to do because you start seeing everything about Lemonade and all the relationships, with the labels scrubbed, it's a little bit less engaging. But it wasn't brought together for my entertainment. It wasn't actually for any human. This is for our agents. This is what enables them to do so much of what you'll be able to see shortly. I want to explain why this is visually pleasing, but it is strategically, foundationally important. Dostoevsky said that it takes more than intelligence to act intelligently. It takes more than intelligence to act intelligently. Now, consider, we've got a smart room of people, a smart audience at home.
Imagine if I asked you to prepare a detailed operational plan for Lemonade's 2027 business. You would struggle to do that, but not for lack of intelligence. You would struggle to do that for lack of context, lack of information. You would want to know a bunch of things, right? You'd want to know how many engineers we have, what they're working on. You'd want to know everything about our code base, what its capabilities are, its APIs, its MCPs. You'd want to know what systems we have today and what is the roadmap looking like. You'd want to know a whole lot of financial information. What is our CAC, conversion, retention, loss ratio, and price elasticity by product, by geography? You'd want to know not just a snapshot, you'd want to know the trend lines too, and you'd want to know a whole bunch of external context.
You'd want to be able to estimate inflation and investment returns, which are highly impactful on our business. Ideally, you'd want to have access to board decisions and management presentations in order to be able to present a plan for 2027 that coheres with everything that's come before. The visualization behind me is of all of that and then some. Lemonade OS has all of that context, every line of code, every specification, every deck, every project, every metric associated with every product. It isn't static. An army of agents are scouring through the context, finding all the stuff, sifting through the detritus, and taking out the trash, and another army of agents is seeking out lacunas, gaps in the knowledge base, and filling it, up to the point of sending Slack messages to us on management saying, "I'm missing this information. Can you share this presentation?
Can you answer this question?" Why should you care? Why do I care? Because, my friend, in the agentic age, in the age of AI, context is king. It is what makes intelligence act intelligently. When, as is the case for most companies, intelligence or context lives in heads, in protein, it is fallible, it is slow, it is horribly siloed, and most importantly, is entirely inaccessible to AI. No company, I put it to you, can harness truly the power of AI without having something like this at its core, at its very foundation. So much of what you will see this morning is downstream of this. The first downstream stop is how we turn all of that context into code, and the answer is continuously and autonomously. What you're looking at is our harness. This is real, this is live.
I'm going to interact with it in a second. This is actually what it looks like. If we conceptualize our context as a neural network, we think of our software as cells and organs that we build, the proteins. As I say, this isn't a mock-up. Each cell here represents an actual Lemonade service, and the molecules within it represent a module that is being worked on. What we're seeing is the Lemonade OS repairing its own cells and growing new capabilities non-stop, 24/7. The OS, that stardust that we saw before, that context, it monitors performance issues, health data, customer feedback. It takes all the information, orders it, and creates a queue of improvements that the software would benefit from, and then it executes on them. It has, as we saw, total context.
You see down on the left side, the queue, my right, sorry, the queue of different projects. I can click on them, dive into them. These tendrils are connections that it's identified, that though it's working on one service, it knows that it is connected to another. So if it changes, the one has to change the other. It understands the relationships between the different modules within, and it looks like it's just completed that one. It's moving on to the next one. It's mesmerizing. I could look at it all day. Friends, back in 2017, Shai wrote two blog posts. One was entitled "The Rise of the Autonomous Organization," and the second, "Don't Hire More People, Build Smarter AI." It's very au fait, very of the moment, but he wrote it in 2017, almost 10 years ago.
Behold the autonomous organization, one that identifies what needs to be built and then builds it while we sleep. There is one more foundational part of Lemonade OS I want you to be aware of, and that is our closed-loop system. For every single prospect who comes to Lemonade, a gaggle of some 50 AIs make a series of predictions. They predict whether and what the person will buy, whether, when, and how much they'll claim, the likelihood to cross-buy something else, to churn, and so on. They then reduce all of those predictions to a real-time estimate of lifetime value, what we expect or predict this relationship to be worth in dollar terms over its duration. Friends, Lemonade OS has made billions of such predictions to date. It adds millions more every single week, hundreds of millions every single year.
Because it is a single coherent system, it is able to watch, it is able to wait, and it is able to close the loop. It compares what it predicted to what actually happened, and it uses those observations to improve the next prediction. Predict, observe, update. So core to how AI works and so alien to how insurance companies operate. That cycle runs continuously, becoming ever more prescient with every turn. Tim, in his section, will show you just how predictive, how precise these models have become. All this talk of predictions brings me to my final point, financial predictions. So let's switch gears a second. You are here because you are all in the prediction game as well. You are also, as we are, in the compounding game.
If you want to divine what Lemonade's next two, four, or six years will look like, if you want to underwrite our path to $10 billion in premium in IFP and a billion dollars in EBITDA, the best way I put it to you is to understand the mechanics that brought us here and play them forward. Equipping you to do that is what this morning is all about, which brings me to the numbers. 2026 is expected to be our first full year of GAAP operating cash flow positivity. By other measures of cash flow, we've been positive for a while, but first full year of GAAP operating cash flow positivity. The current quarter is expected to be our first quarter of EBITDA positive, and we expect 2027 to be our first full year of EBITDA positive.
We expect the year after that, 2028, to be our first full year of net income positive, of EPS positive. So in three successive years, we expect to move from cash flow positive to EBITDA positive to EPS positive, and Nick will return to this in his section in a couple of presentations' time and show you how the one is really inexorably linked to the other. They just flow one from the other. Throughout and into the foreseeable future, we expect to sustain over 30% compounded annual growth. That means that, give or take a few months, we expect our business to continue to double every roughly two and a half years. Double and double and double.
You play that movie forward, and you'll see that by 2030, we're expecting to be about $5 billion of premium and closing in on half a billion dollars of EBITDA, not quite. And two or three years after that, in 2032, 2033, we expect to pass $10 billion of IFP and a billion dollars of EBITDA. So over the next six years, give or take, we expect to go from a billion in revenue to a billion in EBITDA. It's natural to discount or even ignore forecasts like this two, four, six years hence, and you will do whatever you deem best. But let me say this. Lemonade is now an entirely different company than we were six years ago, over six years ago now, at the time of our IPO. Our IFP is more than 10x bigger. Our gross profit is more than 20-fold bigger.
Yet our shape and scale today were foretold by our model, first built in 2019 and submitted to Goldman Sachs in 2020. In fact, we expect our 2026 revenue and gross earned premium to be roughly 10-15 points ahead of the model we shared with our bankers at the time of our IPO, while EBITDA and headcount are expected to be within five points of what we modeled almost seven years ago. I put it to you that something about Lemonade is structurally different. In the intervening years, we've given an updated guidance on IFP and EBITDA about 70x . We have not missed guidance once. We have not lowered guidance once. And it's not that we beat guidance every time by sandbagging. Our median beat was under 4%. There is something structurally different about Lemonade that makes it so predictable, almost mechanical.
Understand the mechanics, play them forward, and you'll have a much better place to make our forecasts investable, to assess their credence. That's what this morning is all about. First up will be our COO, Adina, who will host four demos from different executives spanning claims to pricing to product and marketing. I have tried to show you how the Lemonade OS builds massive contextual self-improving intelligence. Adina's segments will show you how that intelligence is piped into our operations, manifesting throughout our business. Maya, our Chief Business Officer, will then show how agentic operations that Adina shows translate into compounding business growth and profitability, focusing on Lemonade Car. In our final presentation, Tim and Nick, our outgoing and incoming CFOs, will show how the totality of what you see this morning is set to become financially compounding returns in the years ahead.
At the end of the morning, I will return with some parting thoughts. Adina, over to you.
It's so great to see you all again. Daniel just showed you the beating heart of Lemonade OS, a pervasive intelligence embedded throughout our operating system, not superficially bolted on top. Maybe for some of you, that seems like science fiction, or maybe just a little abstract. Either way, my goal today is to show you how that intelligence manifests concretely throughout our operations, and why what you'll see is so important to the Lemonade investment thesis. At any Investor Day you attend these days, companies are talking about deploying AI models. But talk is cheap. We will show you capabilities no other insurer has, driving dramatic business results no other insurer has shown, because access to GPT or Claude or DeepSeek is barely table stakes these days. Intelligence without context is useless, and without tools is powerless.
What that intelligence can see and do autonomously is what matters, and that is a function of your operational plumbing. You see, AI is amazing, but it isn't magical. If you're a legacy company with hundreds of disjointed systems, your agents are going to be as restricted as your people are. Then add your organizational reality, departmental politics, job security, and a culture where technology is somebody else's problem. Even the smartest AI isn't going to fix these leaky pipes. It inherits those constraints. It could summarize a claim or draft a campaign, which is super useful, but it cannot see across the business, take action, and learn from the outcome at scale.
You can try to hire a really smart person to solve this, but I think we all know it doesn't work this way, because if the systems and the data and the processes are a mess, this person may get a few wins here or there, but odds are they're going to want to eventually start pulling their hair out. Fragmented systems create fragmented intelligence. Lemonade entered the agentic era from a very different place. For 10 years, we've built one connected operating system that can see the customer, the policy, and the financials all together. The intelligence doesn't struggle to reconstruct its understanding of the business every time it's asked to do something, and then it can use the same tools that our teams use to act on what it finds, and then it observes the outcome, ingests new insights, and improves the next decision.
Now, you are about to see an in-depth demo where a signal that surfaces in claims becomes an input for pricing. The pricing decision becomes a product change. The product change becomes a marketing campaign, and that campaign sends insights back into the system, a closed optimization loop running continuously. Now, we have especially chosen an obscure use case, something narrow when taken individually. But if you can do a thousand turns, you start to show structural compounding advantage. The speed of iteration is what matters. Infinite small changes drive big impact. Four of our executive leaders will take you through the Lemonade OS working in their respective domains, but a few practical comments before we begin. Agentic work sometimes happens too quickly or without enough visible clicks to make for a useful demonstration.
At certain points, we have added a visual layer or simplified a highly technical flow so that you can get a good appreciation of what is going on. We have also sped up or pre-populated some of the longer flows so that you do not wait while our agents are executing or thinking on some of the more complex tasks. The numbers and the metrics that you see here in the demo are solely illustrative of how the product works. They do not reconcile to our reported financials, so please do not use them in your models. Thank you. The capabilities that you will see here range from technology already in production to active development launching soon. One more thing, this is not just sexy tech.
At the end, I will come back with real numbers and metrics that show where the impact of this technology materializes in our cost structure, our pricing, and our growth. Let us get going. First up, claims. Paying claims is our business. We settle over a million claims every year. Success is doing it faster, better, more precisely. That is how we delight our customers and show up for them in their time of need. To take you through our operations, I am honored to introduce you to Sean Burgess, our Chief Claims Officer.
I am so excited to show you how humans like me interact with Lemonade OS. I have been in this business for over 30 years. This is the most exciting thing I have seen by far. It is actually why I came to Lemonade, to work for a company that pushes the boundaries and capabilities of AI. Let us take a look. This is how I get to start my day, with Lemonade OS giving me a view of the health of the operation. This provides all the details from an end-to-end claims experience around service, quality, financials, many other metrics. But you may be wondering, "Sean, why would you show me a dashboard?" This is not a dashboard. This is a command center powered by Lemonade OS that identifies opportunities and provides solutions in real time. I am going to walk you through an example.
So in this example, we have median settle time, which is red, which is our time to handle claims end-to-end. I can investigate this, and Lemonade OS understands our claims operations, understands our workflows, and starts to analyze what is driving an increase in cycle time. You can clearly see here that it is our homeowner line of business. So the other products are in line. It is the home product. Then within home, we have loss of use claims that have increased. So, loss of use claims are when customers have to be out of their house for things like wildfires or tornadoes, and they need additional living expenses. Lemonade OS obviously goes farther than that in the solution part I was talking about. It gives a recommendation to automate some of the manual tasks that are being done right now by our adjusters.
This is not new to Lemonade. AI Jim has been handling claims for over a decade for us. Daniel mentioned in his opening that we famously handled a claim in three seconds 10 years ago. In the 10 years since then, our claims agent has gotten smarter and smarter and more capable of handling complex claims. So, let us start building this agent. Lemonade OS understands the operation, so it predefines and prefills the information that is going to be needed and the details to create this agent. I have an opportunity to review it in whatever level of detail I deem appropriate. I am going to go ahead and share a few things with you, like the standard operating procedures, that Lemonade OS inputs. These are the same standard operating procedures that we train our human adjusters on. So our AI and our human capabilities are being mirrored.
We have the ability to work with personal assistants, personal agents, as you see here. Again, to me, another example of being on the leading edge that Lemonade is. So let us go ahead and I am good with this. We are going to let Lemonade OS start building. So it looks at the material, it looks at all of the workflows, it looks at other agents that are already operating, and it builds what the agent needs to look like to handle claims and the capability. So, a very detailed workflow, a litany of definitions and scope of what could materialize during a claim. All loss of use claims are not the same. Same coverage, but not the same. So what could happen during that time, the agent needs to know.
This next phase is really important because we have to evaluate our agent against our compliance, accuracy, service standards. One way we could do that would be to audit claims that come in, maybe even audit older claims, and then make incremental improvement to the agent. That is too slow. This is a fascinating part of Lemonade OS, among many fascinating parts. Lemonade OS then has the ability to create a dataset of past claims and workflows and experiences that look like loss of use, that are loss of use, that the agent will need to know, and then evaluate this agent against that criteria, rapidly increasing our proficiency, our speed to proficiency, and scale. By the way, I loaded this earlier, so I am going to go ahead and load some cached data for demo purposes. So this agent is operating at a 98% accuracy.
That is slightly better, actually, than what our human agents are operating at 95%. This agent is ready to be launched into production. I am going to do that. Before I hit that launch button, it is going to take us to our live claim activity sheet I wanted to share with you. It will be busy because that is the point. There is a lot of work taking place. You will see humans and agents that are handling tasks, handling claims. Let us go there. This is the live claim activity field, so there is all kinds of activity taking place on all products, humans, and agents. I am going to find the agent that we just created. Here it is, loss of use agent. This is the claim that this agent was assigned. Actually, it has already handled this claim, but I will get there. This is a four-year customer of Lemonade, Rob.
It looks like he is out of his home because a storm knocked out his power. He has some minor damages. Interestingly enough, he had a home battery that allowed him to stay a few extra days in his home, but eventually that battery ran out. He had to take his family and his pets out, and that is where Lemonade comes in to take care of him. On the left-hand side, you see all the green check marks. That means the agent has already handled the claim. It has done a lot of work, extracted the receipts, it verified the facts of loss, it computed a payment. I am not going to click on all of these, but you get the point. There is a lot of detail. Everything the agent does is tracked here, and then it issued the payment. Rob and his family are taken care of quickly.
Listen, how amazing was that? Lemonade OS, I think it is amazing. Hopefully, you do. Lemonade OS identified a metric that wasn't performing optimally, provided a solution, and then this is probably the more amazing part. Me, as a non-engineer, I was able to create, evaluate, and deploy into production an AI agent. Certainly something I didn't think I would ever do in my career. A small but powerful example of efficiency, accuracy, and flat-out world-class service that is provided by Lemonade OS. Thank you.
Wow. Thanks so much, Sean. Let us recap what just happened here. Sean presented a view of the entire claims operation, then identified a specific constraint related to slowing down of our median settle time in loss of use claims. Then he created a new skill for our claims agent to automate this process. He then launched the agent, making it available both to our internal OS and to our customer's personal agent so that they too can activate it autonomously. Lemonade can expand automation into increasingly complex claims without reducing the control standard. The agent trained on our own structured claims history data, and then its precision and performance was evaluated against a human baseline before touching a live claim. This is how autonomy reaches the loss adjustment expense line without creating any hidden quality or compliance risk, but more on those numbers later.
Loss of use may seem like a narrow example, and it is in isolation, but here's the thing at operating at near 70% claims automation rate. When you are already automating the majority of the claims, the additional work is naturally long tail, but these small things add up to a big impact. At a traditional carrier, the cost of coordinating data, operations, and engineering can exceed the value of fixing any one of them. Lemonade's unified operating system changes that equation. In aggregate, these changes lower variable cost, improve service, and generate more precise data for pricing, and that's where our loop moves next. Pricing is how we outperform the market. Ori Hanani, our SVP Insurance, will take a signal identified by loss of use claims agent and turn it into a pricing decision in our book of business. Ori, the stage is yours.
Thanks, Adina. One of the things I love most about working at Lemonade is how seriously we take defining and measuring success. Get that right, and you can align your people and your AI agents around the same goals and understand what moves the needle. Our framework of choice is OKRs, objectives and key results. Here at the top, you can see Lemonade's company-level OKRs. They define what we are trying to achieve as a business. As I follow these goals, you can see how they connect to the goals of the different product lines, and from there, to the individual initiatives. One initiative can support several goals. One goal can depend on several initiatives. As I follow these connections up and down, I can see what's on track, what's falling behind, and how it affects our company goals. Think about what this means.
Lemonade OS understands our business and our goals. These connections give our people and our AI agents context to act on. So, an AI agent understands why the work matters, what else it affects, and how to prioritize it based on the expected impact on the whole business. Okay. Drilling down to the property business, one of our objectives is profitable growth, and key to that is making our pricing more accurate. That means pricing each home for its own risk rather than the average of a broad bucket. The agent has identified two opportunities to improve on this metric, and one connects to Sean's earlier example. Remember Rob with the battery that helped his family stay at home during an outage?
Intuitively, it makes sense to me that if Rob's battery does half the job and lowers what we expect to pay in claims, that could also mean a better price for Rob. In fact, the agent identified a pattern in the data showing fewer outage-related claims after catastrophes, and the agent points us to a potential opportunity around home batteries. So, the agent identified the signal, ran the analysis, and created this presentation for me. This presentation has all the findings, the evidence, and the limitations. It takes thousands of inputs, analyzes at machine speed, and manifests it in a presentation so that us humans can comprehend the analysis easily and make decisions efficiently. So, five minutes into my morning, coffee in hand, and I'm reviewing opportunities the agent identified and analyzed overnight.
Okay, now in this specific case, the agent concludes that the signal is promising, but we have too little claims experience on our own book. In other words, we cannot yet use our internal claims data to support filing a discount based on this signal alone. I will go ahead and ask the agent for a little more input into this point. Can you help me find the evidence we would need for regulatory approval? Insurers can build on treatments that regulators already approved in competitors' filings, and here, the agent understands what we are trying to achieve and searches competitors' filings automatically. It found an interesting connection. We started with home batteries, and the agent identified an established discount in the market for old-fashioned generators. The logic is simple. Both can keep essential systems running during an outage and help prevent damages.
The agent proposes extending this decades-old concept to modern protective devices like home batteries and Powerwalls. I will take it. Now, what types of discounts could we offer customers? Here, the agent proposes three discount options with an actuarial analysis behind each one. The agent automatically runs all the models and analysis that an actuary would usually do manually. Essentially, everything needed for an actuarial review, evaluation, and sign-off, including the projected impact on loss ratio, the underlying assumptions, and sensitivity testing. Okay, so I reviewed the analysis and the evidence, and I am comfortable with the recommendation. My decision and all the supporting evidence now become a structured build task, and my review is complete. Not to brag, but this is one of my favorite parts of how we built our entire platform. This whole process would have taken my team weeks to complete.
In fact, my team would probably not even consider acting on such a signal simply because our time is limited. This completely transforms the operating model, and it changes what we, as humans, are responsible for. It infinitely extends our ability to get more and more granular in our analysis, and so more accurate in our pricing. This is how you win in insurance. Thank you.
What Ori just showed you is what it looks like when AI turns an observation into a decision. It all starts with OKRs, objectives and key results. OKRs are part of our operating system. They are how we prioritize, focus, and orient the whole company, both humans and technology, around common goals with autonomy. AI and human expertise work together as genuine collaborators. The system surfaced a pattern that could help Ori exceed his goals. Homes with backup power appear to produce lower loss of use claims cost. It assembled the internal evidence, searched the regulatory precedent, and modeled the gross profit trade-offs at machine speed. Now, think about what changed for Ori. His time was not spent chasing teams or waiting for an analysis. Lemonade OS created the choices, whereas Ori made the decision. Eligible customers should receive a new discount on their homeowner's policy.
But now that decision needs to become a product. We need to change the rating factors, the underwriting rules. The customer flow needs to identify who qualifies, and the filing needs to be prepared for regulators. At most carriers, that means months of trade-offs. The speed in which you can ship changes directly into the hands of customers drives the financial impact of those changes faster. Ohad Kedar, our VP Product, will show you the same decision moving directly into production. Ohad?
Hey, everyone. I'm going to show you what a day in the life of a Lemonade product manager looks like. This is Lemonade OS project board, tracking all of our product work. Most of it is handled autonomously by our product agents. I'd like to take a moment and introduce our agents to you, because they're really an important part of our team. Our agents conduct research, drill into data, design customer experiences, and so on. Each one of them specializes in their own domain, and together, they act as a team. Only once they reach something that is critical or calls for our review, they stop, and then we get in. We can see here under the build queue that Ori's task just landed. Ori made a decision. We're going to take this decision and deliver it to production. Let's dive in.
We can see the agent has come up with a pretty thorough implementation plan. Initially, hearing about that discount sounds like a fairly simple change. In reality, this change calls for a lot of moving parts behind the scenes. We need to prepare the filing to send to regulators. We need to implement rating factors, change underwriting rules, add customer-facing questions, and so on. Conventionally, for incumbents, these kind of changes would require multiple teams and months to complete. We are going to complete it in this session. We're going to start with filing for two reasons. First, it really is a long and manual process. More importantly, it's not that uncommon to be getting objections from regulators. When that happens, back and forth begins, and timelines really start to move slowly.
Imagine we have a crystal ball that could help us predict what kind of objections we could get. We don't need to imagine. Lemonade OS is trained across all filings from all carriers in all states and product lines to predict what kind of objections we could potentially get. This process could take a little bit of time, so I'm going to load cache results to speed things up. Okay. I see that for this particular filing, we have low risk, so I feel confident moving ahead and start generating these exhibits. The agent will now prepare everything we need to send over. Rating exhibits, policy forms, actuarial memorandum. As we can see, there are many of these documents. Some of them last over 500 pages, including legal precedents, historical data, and datasets from multiple states.
We can also see that this would still call for the filing team approval, considering how sensitive these filings are. That's okay. We can keep on going while the team reviews. Remember that Adina mentioned we're going to slow the system down so we can see the agents in action? I'm actually going to get back to normal mode and speed things up, so don't get alarmed if you can't read everything you're going to see on the screen. The agent will now take care of all these technical changes that need to happen. It will change these rating factors. It will update our underwriting rules so only eligible customers can get that discount. Only if it reaches something that is critical or calls for review, it will stop. It just did.
There is a customer-facing change that needs to happen, and we want to make sure our customer experience remains in the top quality possible. Let's review the plan. I see the agent wishes to change our home flow and to ask the customer if they have a backup power system installed at home, and the customer will be able to choose between a generator and a few different battery options. That plan is okay, but we can do better. Here's my thinking. Some EV car customers could potentially get a better bundle discount with this, so let's change the plan. Let's add this to the car onboarding flow. The agent will now go ahead and change its original implementation plan to also include it in our car product. This plan is solid, and we can move ahead and implement the changes.
The agent will now implement the changes, but it would still not commit any of the changes. It would first show me what these changes would look like in our product. This is our mobile app. This is what our customers would see over their phones. By the way, it's also what their personal agents would see. Anyhow, I think that question looks great, and we can move ahead and apply these changes. With that, our implementation plan is completed. Just before we launch, I can see that this is going to touch about 23,000 of our customers. Here's everything that's changing, and we just got that approval from the filing team, so without further ado, let's hit that launch button. This is now being sent to regulators, and boom, this feature has shipped. Thank you.
Boom. Boom. Ohad just moved a complex change across two products into production, factoring in both how humans and their personal agents will interact with it. In fact, that's what we did in reality. This discount went live about a month ago. There are states in which we can file and launch concurrently, and others in which we have to hold for approval as our product releases gradually extend nationwide. This gradual rollout allows us to observe how it performs. Did the customers convert as expected? Are gross profit and LTV tracking to plan? How should the growth team respond? Before we do that, I want to give you one more observation to ponder on, leading up to the last segment of our loop. In many AI workflows, humans continuously prompt and assign and check on the machine.
In the demos that you've seen today, the system works within defined authority and routes the exceptional decisions back to the human. It's the human attention that becomes the scarce resource that the OS allocates deliberately. As COO, that has deep consequences in how I think about our operating model. It changes team structure, management cadence, and of course, the amount of work a small group can supervise, as working rhythm constantly evolve for agentic operations. Our growth team wakes up to a floor of agents that have worked throughout the night without waiting for any brief or approval. They've made budget decisions, optimized creative, all while the team was asleep. Michael Bloch, our VP Revenue, will close the loop and show you how we turn product learning into growth. Michael.
Everyone, you've seen some amazing capabilities today across claims, pricing, and product. I want to show you how they all come together on the revenue side. What you can see here at the top are our core growth metrics, starting from marketing spend on the left all the way to in-force premium on the right. Below them, you can see what I like to call the production floor, where you can see all the decisions and actions made by our agents across the growth team. Agents themselves aren't really the interesting part anymore. Everyone uses agents. What's cool about how we've implemented them is that they're connected to the same system you've already seen today, and that enables them to make bigger, better-informed decisions and to act on them autonomously. There's no better place to see that in action than budget allocation.
What you can see here are all the budget allocation decisions made by our agent over the last 24 hours. The reason I wanted to highlight this is that budget allocation is really hard. You need to know how every channel is performing, where the opportunities are, and where our next dollar should go. Then you need to move that money quickly across many channels and partners. We operate in a highly competitive industry where acquisition costs move fast, so being able to react and redeploy spend is a real advantage. Doing that 18 times overnight with significant business impact without needing me involved, that's pretty mind-blowing. Speed is only part of the story. Let's take a look at what went into one of these decisions. Let's look at the back-up power cohort decision and close the loop on that signal that we saw earlier.
Here is the decision screen. Let's start on the left. What triggered this decision was strong performance of the back-up power cohort, particularly in lifetime value relative to acquisition cost. The agent evaluated multiple scenarios and eventually chose the one it thought would deliver the best outcome. What's interesting here is that a marketing budget decision is being made based on our financial objectives, not marketing metrics. As a marketeer, that's pretty crazy. You usually don't get this type of connection in real time. To understand how this is even possible, let's take a quick tour of the brain behind this decision. What I just called the brain is in fact a Lemonade OS agent that predicts with high confidence what our marketing spend is going to produce, which you can see in the financial metrics at the top.
Normally, to do this, you would have to pull forecasts, performance data, and insights from across the business. Even after all of that work, you may end up with two or three different scenarios. Here, the agent effectively maps out the full range of outcomes on this curve. Then, based on the business objectives and constraints we give it finds the optimal point for us, which brings me to the fun part. We can change these. You can see the plan is optimizing towards gross profit right now, but what if we wanted to lean harder into growth? I am going to change the objective to revenue, and also, I am going to lower our LTV to CAC threshold just a little bit and allowing us more room to grow. Let us simulate. Now we have a new plan.
The point on the curve has moved, and the financial objectives at the top have shifted as well. That is the part I find pretty remarkable. A change in the business objective that would normally trigger a whole new planning process across the company gave us a new plan in seconds. In this case, I have to admit, I actually prefer the plan that the agent originally recommended, so I am going to reset it. Great for Lemonade, slightly less reassuring for my own job security. Let us go back to that decision, because so far we have only seen how it was made. If we scroll down, we can see everything that this one single decision has set in motion. Budgets were moved across channels, new creatives were deployed, partner opportunities were surfaced, each picked up by the relevant agent.
There is one more decision I wanted to highlight that connects back to what Ohad just showed you earlier. Ohad's back-up power question gave us a new signal. As the data came in, one segment started to stand out, Tesla owners with Powerwall. That triggered the concept and ideation for a new campaign. This campaign is already live, by the way. More budgets are moving against it, and the results are already feeding the next decision. This ability to close the loop fast is what separates a good acquisition machine from a great one. Honestly, I would put ours up against any direct-to-consumer business out there. Thank you.
What you just saw was one decision traveling through the entire insurance value chain. Claims produced the signal, pricing converted it into a defensible decision, product shipped it, growth put it in front of the right customers, the outcome returns to the system, and the next iteration begins with the information the previous one created, and again, and again, and again. Every insurer learns from its data. The difference is latency. Because insight and decision and execution all sit in the same system, Lemonade can complete countless iterations every day, each turn of the crank making the next one slightly better, faster, cheaper, minor in every turn, but overwhelming across thousands of them. Okay, so we have got the tech. What is the operational impact? Let us connect the OS to the financials.
Starting with claims, our LAE ratio, which is our loss adjustment expense as a percentage of premium, is at 5%, leading the industry at nearly twice the efficiency of incumbents despite being a fraction of their size. LAE means the cost associated with handling, investigating, and settling the claims, separate from the claims payout itself. It basically means the operational overhead of handling a claim. Every point we remove allows more premium dollars to reach the bottom line. 80% of our customer service interactions are fully automated. This means the vast majorities of inquiries from our customers and their personal agents, such as policy updates and address changes and interested party additions, are handled by the Lemonade OS. Note the trajectory of this graph.
Our automation rate sustains a healthy increase, evolving our customer operations departments from otherwise variable cost into fixed ones that scale as our premium continues to grow. The next number sounds made up. I understand that. It's not an aspiration or a projection. It's just simply what our internal measurement shows today. 99% of our new code is written by AI. As Daniel showed you, engineering tasks are increasingly generated and executed autonomously by our OS. That changes the amount of product functionality each engineer can ship, but also the speed at which ideas can reach customers. We've developed a new internal AI multiplier metric across all corporate functions. It measures the throughput and the velocity that teams can produce without adding any additional headcount, from compliance and finance to actuarial, product design.
Our corporate teams are performing at twice the velocity they had been a couple years ago. To be honest, this one's a bit close to heart, because we're proud to see measurable impacts from AI development across all departments, not just the obvious ones. These four metrics measure different kinds of work, but they all point to the same economic change. Premium and revenue grew faster than the operating base required to support them. Putting all these stats together, and set aside the enormity of the value created, efficiency gains have saved us $400 million in personnel expense in just the last four years. To bring this point home bluntly, let's take a look at three charts, quite familiar, overlaid one on top of another. IFP grew nearly 3x , and at the same time, our customer count nearly doubled.
Our employee base has stayed basically flat. What I just said is astoundingly unusual in insurance. The industry has always grown through a fairly predictable equation. More policies require more service work. More claims require more adjusters. More products and states add underwriting, actuarial, and compliance work. Our operating system, which you've just seen, gives us the ability to rewrite that equation. Talk about COO life goals. Technology absorbs an increasing share of the work while enabling each person to oversee far more output. Growth continues without headcount tracing the same curve, and we expect this trend to continue. This gives us more than just a lower cost base. This capacity that we create, it's ours to deploy. We can use it to lower our prices or expand our margins or pursue opportunities that would either be too expensive to justify. Let's look at our marketing efficiency.
Our LTV-to-CAC ratio is held at our target average return of three. Now, look at our marketing spend. It has increased consistently in absolute terms over the same period of time. Normally, one of those lines eventually gives. As companies spend more, they exhaust the obvious opportunities, and each additional dollar works a little less hard. They can preserve efficiency by limiting growth, or they can keep accelerating, but in that case, they will be accepting declining returns. Technology expands the opportunity set. We can see where the next dollar should go, what we expect it to return, and whether reality matched our prediction. That makes any decision to invest a grounded one. The remarkable part of this chart is not simply that spend is rising, it is that we are putting materially more capital to work without diluting the unit economics, and that is a gravity-defying dynamic.
Maybe we are just going after customers that nobody wants. That brings me to our last chart, which demonstrates the impact of AI-driven pricing. IFP has been growing at above 30% year-over-year, and loss ratio is down meaningfully. Now, let us add conversion. It has held during the same period of time. I think we all know how difficult that combination is. Improving loss ratio usually means tightening the aperture and turning away more customers. Increasing conversion and growth usually means opening it wider, but living with the consequences later. Lowering loss ratio while maintaining conversion usually means shrinking to excellence. Technology changes this paradigm. It does not have to be a deadlock. The system that you just saw can distinguish between customers who may look similar on the service, predict their lifetime economics, and identify the price in which the right customer is likely to convert.
That precision allows us to be more selective and more attractive to our customers at the same time, as constant de-averaging chips away at monolithic prices. That is the compounding in action. Growth up, loss ratio down, and conversion holds. That chart is the clearest evidence that I can give you that our technology is not sitting alongside the business. It is changing what the business is capable of doing, and there is your smoking gun. Today, you saw the inner workings of Lemonade OS. The demo showed you how context moves with every decision and how each customer interaction helps the system make the next decision better. The numbers show the direct financial impact on our cost structure, our pricing, and our growth. AI is completely transforming our industry, but we believe Lemonade is uniquely positioned to lead the future of insurance.
Next up, Maya is going to talk to you about how we are putting all that tech to work in our car product, where the next chapter of growth is being written. Thank you so much for your time.
Thank you, Adina. You have now seen how our technology works, how the different pieces connect, and what makes Lemonade fundamentally different. I want to take this one step further. What does all of this make possible? When we can serve customers more efficiently and price risk more precisely, how much more room does that give us to grow? That is what I want to spend our time together this morning, turning that technological power that you have seen today into something very tangible, profitable growth. Nowhere is that opportunity bigger than car. With an approximate market size of $400 billion in the U.S. alone, car insurance represents the biggest growth opportunity for Lemonade today. I am happy to say that in every single account that matters, we are outperforming our previous expectations.
With IFP growth of 63% year-over-year, almost double the growth of our book, very healthy loss ratio, and almost half of its IFP coming from existing customers, we are bullish on our ability to capture market share in this market. These results give us confidence in how big car can become, and that confidence comes from experience because we have already seen this model, what it can do in renters and in pet. Both of them started from a small base, but let me show you where they are today. Our renters and pet product continue to grab increasing market share, with close to 10% nationwide market share in renters and close to 9% market share in pet that launched four years later. To put that in perspective, Allstate has been providing renters insurance for more than 65 years.
We launched a decade ago and have a larger market share. 10% market share in renters is great, but it understates things. We sold our first policy a mere 10 years ago, and we achieved nationwide availability only a few weeks ago. In other words, our competitors built up their market share during the years that we were not even in market. Maybe a more telling metric is looking at Lemonade's share of policy growth in the market. According to best estimates using industry data, Lemonade's market share of policy growth in the category is about 50%. Half of the market's growth in policies is attributed to Lemonade, the same as the rest of the industry combined. In pet, where Lemonade is not yet even available nationwide, that number is 40%, just one small step behind.
I would like all of us to squint together so that we can take a look at Lemonade's current market share in car. We have a 0.1% market share. It is tiny. Let us do the math. If we do half as well as we already are today at renters at 5% market share, this represents a $20 billion opportunity in car alone. I am not suggesting that car is simply going to follow the exact same curve. It is a different market with different dynamics, and it is highly competitive. The same ingredients that helped us win in those markets are very much present in car, and even more so. We are also seeing similar dynamics and metrics that we have seen in renters and in pet at the exact same time point.
I believe that after I will share with you our numbers and performance today, you will feel the same. Let us start with the big question. What does it take to win a significant market share in car? The answer is actually pretty simple. It is all about the price, price. My colleagues will remind me that having a compelling brand with best-in-class experience and superior service are also very important, and they are. But we have already proven that we can do that, and in fact, we have redefined NPS in our industry. Price is the leading consideration when customers choose for car insurance. The question now becomes, how can we deliver a better price and attractive economics at the same time? The answer to that is also rather simple. It is all about the cost, cost.
Lemonade has a significant structural advantage that only grows with every new customer we bring in. And we have not one, but three cost advantages. Our cost to acquire, our cost to serve, and our cost of risk. The first is our cost to acquire. We have built a brand and a growth machine that brings in young customers at a low cost of acquisition. You have seen the numbers of pet and renters. Just this year, we have already generated more than $100 million of new car sales from our existing customers, with zero cost to acquire. This is no longer a footnote in our car sales, but a significant needle-moving number. These customers also have a lower cost to serve, which is contributing to our second structural advantage, our structural cost to serve.
Adina showed you some of the incredible numbers as we continue on our path to become the most efficient insurance company in the world. And in some of the metrics, we already are. With every new customer, our average cost to serve becomes better, and we have seen a staggering and consistent reduction year-over-year. Over the last three years, we have cut the average cost of serving claims by more than 60%, and customer service costs by more than half. But even in car, with our existing tiny market share, we have reached an LAE cost of handling claims of 6%, which is best in industry. The industry's average is 10%, and they all benefit from efficiency of scale, which we do not yet. This is directly contributing to our gross profit, and we are nowhere close to finish. Our third structural advantage is our cost of risk.
Our system consistently allows us to see and understand risk better than our competitors. We have built a pricing infrastructure that continuously finds new factors and de-averages what might seem like monolithic groups. And with more than 90%, 90% of our car customers using continuous telematics, not a snapshot or in the moment, we see things others cannot, giving us a sharper view of risk. We are going through such a distinct revolution that is directly impacting car insurance. You are all seeing it and feeling it. Having a dynamic and learning system is critical, and this is true for all of our product lines. But let me give you some examples that apply only to car. A parked car carries a completely different risk than a driving one. For the time it spends in the garage, we need to price its exposure, like bodily injury, completely differently.
When someone is driving, well, now it depends who is driving. The risk changes as well. We can create a driver's unique signature after only five drives and differentiate between the different drivers in the household to combine for a more accurate price. But what about when the car is driving itself? Not all autonomous capabilities are created equal. We can differentiate between miles driven by a person or miles driven by different systems. When you go to sleep at night and your car software gets updated, you should wake up with better insurance prices. Lemonade was the first insurance company in the world to offer 50% lower rate for FSD, Full Self-Driving Supervised miles. When Tesla launched FSD Supervised Lite, we quickly followed by offering 30% for FSD Supervised Lite. The ability to bring back savings to our customers is no longer theoretical.
I want to show you a live example of how, as we kept sharpening our pricing precision, we were able to lower prices 10% for select segments. Let me show you what the commercial impact was. We lowered prices by 10%. How did that impact these customers? First, our conversion increased by 50%. Sensitivity to price. The retention went up by 10%. Our loss ratio increased by only 2% since the lower price made us more attractive to great customers. But the most important number that we track and we optimize for is our lifetime gross profit. That went up by 10%. This is the overall profit Lemonade was left with. So even after reducing price and then collecting less premium per customer, our lifetime gross profit is now higher.
With these numbers, we will take this trade all day long and cut prices when we can again and again. I know that we will be able to do this again and again because our costs across the board will continue to go down, helping us push the price down again. This is a flywheel. Lower acquisition, servicing, and loss costs combine into pricing power. Better prices attract more customers, more data sharpens future decisions, and lowers our average cost to serve. We are seeing this with many of our target demographics already. Lemonade is already winning with them. First, with our existing customers. These are customers that have another Lemonade policy that is not car, and they are outperforming our regular conversion rate by 1.5x, 50% higher.
This is because we are able to offer them great prices, and they still have the same loss ratio as the rest of our book. Existing Lemonade customers are a natural, lower-friction car audience for us. We know a lot about them. We do not need to pay to acquire them, and they know and love our brand. This segment is going to continue to grow as we capture more and more market share with our on-ramp products, pet and renters. These customers often belong to another group Lemonade is outperforming with, young drivers. Traditional pricing proxies can disadvantage safer young drivers. I am sure some of you in the room today have had a young driver join your policy, and you have witnessed firsthand how your premiums shot up. Lemonade is able to identify the safer drivers and offer them a better price.
These are strategic customers for Lemonade, bringing the better risk earlier. It might seem small today, but they compound with time. In fact, according to third-party independent raters, we are 22% cheaper on average for them. How do we know they are safer drivers? Easy. Their loss ratio is in line with the rest of our book. We want these drivers to come to Lemonade all day long. What about low-mileage drivers? The vast majority of low-mileage drivers are overpriced by our industry, which creates a huge opportunity for a player like Lemonade with continuous telematics. Self-reported mileage is a highly inaccurate way to price them. Lemonade is 25% cheaper on average for these customers versus the competition, with zero increase to loss ratio results. We are seeing even better results with FSD Supervised drivers, 31% cheaper than competitors. Their loss ratio results? Even better than our book.
If this continues to outperform this way, this means that we can cut prices even more for this segment. What makes this segment so interesting is that the technology keeps improving, and with it, the risk can change. This goes well beyond FSD. Cars across the market already have different levels of driving assistance and automation, from automatic braking to systems that take on more of the driving to prevent accidents. A software update can change how a car drives, and that creates an opportunity for us to understand what changed, measure its impact, and reflect it in the price. As cars do more of the driving, understanding the technology becomes an increasingly important part of understanding the risk. Now look at who Lemonade is winning with. Autonomous cars and young drivers. In other words, with the cars of the future and the customers of the future.
Play this movie forward, and you will understand why we are so bullish. Every new customer makes this flywheel stronger. More data, more efficiency, and better prices for more segments. The technology itself that is powering us keeps getting better, giving us more room to turn that wheel even faster. Now, we have all seen in the room how technology can reshape insurance. First, customers could buy directly and bypass their insurance agent. Then they could just go online and buy insurance without talking to anyone else. Each of these shifts created an opportunity to build insurance businesses that are today worth hundreds of billions of dollars. This is what we have spent the past decade building, agents and the infrastructure that allows them to work together. Nobody here believes that you can just pivot your way into this change. You can't buy AI features and expect to get the same result.
We are already saving customers far more than 15% by bringing back agents, just not the same ones. This playbook will allow us to go after more and more segments in the market. As these agents become more capable, we can solve more problems, cut more costs, bring better prices to more customers. We were built for this moment. If you had to choose one insurance company that would benefit the most from the agentic era, who would you choose? With that, let me hand it over to Tim Bixby, our CFO.
Another Lemonade Investor Day really provides us with a valuable checkpoint. It is a good time to look backward and look forward, and to see how our models, our projections, and our goals have held up. How did we do? How has the value embedded in Lemonade OS and our deep AI optimization translated into growth and profit progress over time? Where we come from, I think, can tell us a fair bit about where we are headed. It also marks a bit of a transition point from a period of consistent execution to one where our advantages have clearly begun to compound. I will briefly cover our past and our present, and then I will hand over to Nick, who will walk through the forward view. Let us cut to the punchline first. Today, we note four core ambitions.
Looking forward, continued high growth, not just maintaining 30%+ top-line growth, but building the conditions to maintain that high growth rate for the long term. With EBITDA positive expected in the current quarter and a full year of positive EBITDA expected in 2027, we also expect the full year of 2028 to be net income positive. Then, beginning to scale both EBITDA and income much as we have scaled premium and revenue to date. The same model and the same approach that enabled our top line to grow from zero to $1.6 billion+ will now enable a path from break-even EBITDA to a billion dollars plus in the coming six or so years. Stepping back a bit, underpinning these ambitions are a lot of valuable tools, as you have seen. Key among them is our sophisticated understanding of customer lifetime value, or LTV.
We can look back at the last three years of our LTV cohort modeling, and the results are impressive. This is the output of billions of predictions Lemonade OS has made to date, millions per week. As Daniel noted, each incremental win is small, but in aggregate over time, the results are clear and the impacts compound. A less than 5% variance between our actual and our expected results during a period where we increased our total spend by 5x. Please recall, nearly all of this was spent on direct customer acquisition efforts and very little on humorous mascots. Here are the results of that LTV model improvement in gross profit terms over the same period. Each quarter, each year, the cohorts layer. Each year's customer acquisition adds a profitable tier on top of those beneath it.
Importantly, the earlier tranches of gross profit not only sustain, but typically grow over time. Each new tier is larger than the one before, generating more gross profit, indicating a greater than 100% gross profit retention rate. If you wanted to choose just one metric that indicates our progress at Lemonade, gross profit just might be the one, with a more than 10x increase in total gross profit to more than $450 million in just four years. Our track record is pretty good. Why should you trust us? We do what we say. The quarterly view of our key metrics as a public company has been right on track 26 quarters running. We have expanded our product offering notably to five products, heading towards all products in all 50 states in the U.S.
We're now live in all 50 states in the U.S. for our renters product, having launched 20 new states just this year. We've begun to mimic that progress in Europe as well. By the way, we drove our pet business in just four years to a size that took one of our leading competitors nearly 11 years to achieve. We've expanded car coverage deeply through telematics, pay-per-mile pricing, and now into fully supervised driving coverage. In Car, launching new states with the most sophisticated autonomous driving coverage is just another day at Lemonade, while most of our established competitors struggle just to begin to understand these seismic pricing shifts. In short, we have the products we need.
Growth acceleration quarter- after- quarter, cash flow positive, European expansion, a dramatic surge in our pace of newly launched states, all with underwriting firmly stabilized, all delivered on or ahead of schedule, and right on track to deliver EBITDA positive in the current quarter, a date unmoved four years after our first commitment and with no need for additional capital along the way. All of this before any material benefit from the onset of the personal agent era, which is now upon us. Most of these accomplishments have occurred in a pretty tumultuous period. Over those same six years, you can see here our stock price volatility, arguably among the least helpful measures of our progress, during a period that held pandemic, war in Europe, a 40-year inflation shock, record cat losses for many players, and notable interest rate volatility.
Yet, the slope of our top line premium, shown here in dark pink, and our bottom line EBITDA margin, shown here in white, tell the underlying story, steadily up and to the right year- after- year. Steady growth and steady margin expansion through the most disruptive macro environment in recent memory. The operating record and the stock chart have rarely been on the same page, and I expect that may change before too long. As Adina laid out, growth rate acceleration and loss ratio improvement in tandem are a rare combination in insurance. Historically, strong players have faced a choice between one or the other, and they seldom coincide. We have been able, over the last four years shown here, to deliver both. A growing top line growth rate coupled with notably improving and stabilizing profitability, roughly a 10x improvement in gross profit in the last four years.
An 80% CAGR sustained across multiple years is unusual at any scale, and exceptional at nearly half a billion dollars. The drivers? Loss ratios have improved. Reinsurance has become more efficient as our data quality has improved, and automation consistently continues to reduce our cost of revenue. The gross profit growth rate likely moderates as the base grows. That's more math than anything else, but the trajectory of the underlying drivers has continued unchanged. You may be familiar with a concept called the Rule of 40. It's been a useful tool or a benchmark, especially in the tech world, in Silicon Valley for some time. The measure is literally just annual top-line growth rate plus profit margin in percentage terms. Historically, when that sum equals or nears or exceeds 40, good things tend to coincide in terms of investor viewpoint, valuation, and market potential.
It is really just one measure of how well a business might be able to balance both growth and efficiency. This balance across both growth and profitability is exactly what creates promise and optionality in a business like Lemonade's. We are currently nearing 35 on this measure, clearly weighted towards growth and clearly headed in the right direction. We expect to approach 40 within our short-term forecast horizon, and this is especially notable given the overall market we are playing in, with an addressable market size, or TAM, hovering somewhere around a trillion dollars plus across the U.S. and the E.U. personal lines markets. Our forward view is all about balancing the two, both growth and profitability, all with the primary aim of maximizing gross profit. For a bit more detail on the forward view, I would like to hand things over to Nick Stead.
Nick has helped build the very financial model that supports this forward view, and in many ways, has helped build Lemonade over the past several years in various roles. Nick steps into the CFO role in the new year, and I am pleased to hand him the stage to walk you through our road ahead.
Thanks, Tim. I wanted to open by acknowledging Tim and thanking him for his leadership. Tim's leadership really shaped our path and brought us to where we are today, and I look forward to continuing to partner with him as he steps into his new role on the board of directors in the new year. Picking up on Tim's discussion of the Rule of 40, I want to first address a choice that sits at the heart of our financial strategy. I will frame this as two scenarios that I will continue to come back to throughout my presentation, and that is whether to prioritize near-term margins and grow at rates similar to our peers, let us call that harvest, or to reinvest to maximize long-term value, let us call that compound.
Both those cases sit at or near the Rule of 40 frontier, so the question is not whether we can be profitable. Rather, it is how aggressively we choose to invest. Here is the thing. Both cases have always been available to us. If we had chosen to prioritize profitability over growth in 2026, we would have generated $100 million of adjusted EBITDA alongside positive net income while growing in line with the market. Instead, we chose faster growth, and I would make the same exact decision today. Let me tell you why. Earlier today, Daniel spoke of the impressive accuracy of the financial model from the time of our IPO. That model has particular significance to me. When I first joined Lemonade in 2019, my very first project was to build it.
Since, I've spent years working with it, challenging its assumptions, and watching those predictions play out over time. That experience gives me tremendous confidence in the underlying drivers of the business and what they tell us about the future. Let me take this model forward by comparing these two cases, compound at our Q4 guided growth rate of 33% and harvest at 10%. Over the next six years, that difference in growth rates drives dramatically different financial outcomes. By 2032, compound reaches $9 billion of in-force premium versus $3 billion under the harvest case. If we compare 2032 adjusted EBITDA, the difference is just as dramatic. Now here's where it gets interesting. If we normalize growth spend in the terminal year, 2032 here, to resemble the harvest case, the underlying profit of the compound case is 4x greater. 4x .
Compounding is the value-maximizing path, and it's not even close. Three key drivers underpin all of that analysis. First, on the growth side, as Maya covered, we will capitalize on the land grab opportunity ahead of us in Car while expanding Car sales to existing Lemonade customers. These CAC-free sales are growing at more than twice our overall growth rate, and as they scale, they are an increasingly powerful driver of growth efficiency. Second, operating leverage drives expanding profitability. Adina showed you how we expect headcount efficiency to continue improving dramatically. Third, we'll maintain strong unit economics as we scale our growth spend, and that's what I want to turn to next. LTV to CAC is the industry gold standard for measuring marketing efficiency, but I find CAC IRR even more powerful. It represents the internal rate of return on our growth spend.
For those who are not in model nerd mode like we are most of the time, it adds the time aspect, so it doesn't just measure how much we earn back on our investment, but also captures how quickly we earn it. Earlier, Michael explained how Lemonade OS determines where the next marketing dollar should go. I'll try to answer a slightly different question, which is, how much should we spend? My orientation and that of our entire leadership team has always been to invest aggressively for as long as the returns justify it. That means continuing to spend when the expected return on the marginal dollar exceeds our financing costs, which is currently 10%, and we balance that against our broader financial goals around growth and profitability.
When averaging the returns generated across our entire budget, this approach produces a CAC IRR of 50% alongside an LTV to CAC ratio of 3:1. My guess is that most folks in this room would consider 50% IRR generally quite strong. But we agree, and we've been able to spend more and more while maintaining that return profile by increasing customer lifetime value and opening new markets and channels. But we're really just getting started. We see tremendous runway to continue to expand investment while maintaining those returns, and that is exactly what we're focused on. With stable and predictable unit economics, we have lots of flexibility and control. We see growth as a dial that we can turn. We can prioritize near-term earnings or accelerate growth as opportunities and market conditions evolve. Let's try a quick live scenario analysis now together.
I'm currently showing you the six-year harvest model, and in the top right, you'll see 2032 adjusted EBITDA is $300 million. In a moment, I'll turn the growth dial to the compound case, and I want you to watch what happens to the adjusted EBITDA margin curve here in the bottom right. At lower growth rates, near-term margins are higher, but as I increase the growth rate, something interesting happens. Over time, the benefits of scale kick in, margins expand, and surpass those under the harvest case. Now, the 2032 profit margin might be just one point higher, but the business is triple the size, and 2032 adjusted EBITDA now nearly $1 billion, and that's the power of compounding. We have clear visibility to sustaining IFP growth above 30% for many years ahead.
That's more than triple the rate at which our competitors are growing, and we're right on track to deliver profitability across all key metrics. We're already cash flow positive, and adjusted EBITDA and net income are inextricably linked to cash flow. Break-even across these measures follows a fairly predictable progression. We continue to expect positive adjusted EBITDA this quarter and for the full year 2027, followed by positive net income for the full year 2028. As we cross into positive adjusted EBITDA, book equity will grow each year from 2026 and onwards. As profits build, we'll generate substantially more capital surplus than we need to support our growth plan. This marks a fundamental turning point. Of all the financial milestones we've discussed today, this is one that I find especially consequential, because it means that we are in control of our own destiny.
Combine a growth rate above 30% with expanding profit margins and a growing capital base. The result? Clear line of sight to excellent return on equity above 30% in the next few years. Let's play the movie forward. Each year, we acquire new cohorts of customers with attractive economics, adding new streams of gross profit on top of those generated by prior cohorts, and that's how successive years of disciplined growth investment compounds into an expanding and predictable earnings base. As Tim showed you, we're a team that lays out bold financial goals and then works hard to deliver against them. We bring that same ambition and discipline to our next set of goals.
They're the culmination of everything you've heard this morning, our differentiated AI capabilities, the opportunity ahead in Car, and the financial model that will take us from a billion dollars of revenue to a billion dollars of positive adjusted EBITDA. As I look ahead to the next chapter of Lemonade, there are three conclusions that drive my absolute conviction in our financial future. First, reduced risk. At our last two Investor Days, we laid out plans for the future that involved significant expected EBITDA burn, and that raised legitimate questions around our path to profitability and the capital we would need to get there. Today, we enter a new phase of sustained profitability, and those profits fund our growth. The risk profile of the business has fundamentally changed for the better. Second, flexibility and control.
When we choose to turn that growth dial, it is because we know that is how to maximize long-term value. Third, the power of compounding. At the heart of our model is a differentiated AI architecture, structural cost advantages, competitive and accurate pricing, more business, and more insights, and that is the engine that drives our industry-leading growth rate. We spent the past decade building and proving out this model. The years ahead will be about realizing its full potential. With that, I would like to invite Daniel back on stage for some closing remarks.
What you just saw this morning was quite different to what you would have seen at any other insurance company's Investor Day anywhere in the world. It was all about technology, AI, and how what Lemonade has built over the last 11 years has positioned us to be singularly efficient, growing, and profitable in the years ahead. In part, this is just the Lemonade machine doing what it does. We built a compounding machine, which we described seven years ago at the time of our IPO, and we have been turning its crank ever since. The machine has delivered everything that we said at IPO and everything that we said at our first Investor Day and everything that we said at our second and at all of our quarterly earnings calls since then. That predictable, bankable performance can make it seem like nothing has changed, just more of the same.
Friends, at some point, a change of degree amounts to a change of kind. I put it to you that Lemonade today is qualitatively different than when we last met two years ago. It is qualitatively different in that Lemonade used to be the high-risk, high-reward insurance play, whereas today we are the low-risk, high-rewards insurance play. There is one obvious change. This year is going to be our first year of GAAP operating cash flow, and hitherto, cash flowed out, henceforth cash flows in. Changes do not get more stark or consequential than that. A business that is generating cash is qualitatively different in terms of its risk than one that is burning cash. One that is generating cash while growing at over 30% is simply without parallel in our sector.
There is simply no other P&C insurance company, to my knowledge, that is showing our kind of growth while throwing off cash. None. Becoming cash flow positive and EBITDA positive imminently, these are step function changes in our invest ability, and I imagine that in that I am preaching to the choir. Cash flow and EBITDA speak to our risk profile relative to our former selves. I want to talk about our risk profile relative to the incumbents, how we are lower risk than the incumbents, how those tables have turned. Here, what has changed is not so much in Lemonade, but in the world. The world is coming to us. In the early 1600s, Galileo Galilei wrote of a thought experiment. He said that a person in the hull of a ship can not tell if they are moving or not. Einstein liked to refer to this 300 years later.
It is simply because when you are in the ship, the immediate surroundings are moving at the same speed that you are. We have all experienced this on a plane. You are on a jumbo jet somewhere, flying at 600 mi per hour. You have a glass of water or coffee or wine in your hand, and everything seems incredibly tranquil, incredibly steady, nonchalant, even as you are zooming away at 600 mi per hour. The explanation is simple. The glass of wine and you and the plane are all moving at 600 mi per hour. That, I think, is a helpful metaphor for where we stand today. We stand at the dawn of the AGI. Everything is accelerating. Everything is changing. The ground beneath our feet is moving, and it is moving fast.
The question for us, the question for you, is who in our industry is able to keep pace, to move at the speed of change even as that speed continuously accelerates? Despite moving orders of magnitude faster than the incumbents, or perhaps because we are moving orders of magnitude faster, inside Lemonade, things feel steady, ordered, smooth. I hope you picked up on that this morning. The world is moving at 600 mi per hour, but so are we. We have been moving at the speed of AI since our inception. A world moving at breakneck speed does not break your neck if you are moving at the same speed. But if you are moving considerably slower, it can be very destabilizing, very dangerous, very risky. The point I am getting at is that the strategic risk in the coming era is obsolescence. It lies squarely with the incumbents.
Lemonade is now the low-risk, high-reward play. But is not every insurance company doing AI? Are not they all making announcements, all deploying systems? Are not they all investing, and do not their budgets dwarf Lemonade's? Is not it just a matter of time until they catch up? Well, yes, their budgets dwarf ours. But no, it is not a matter of time until they catch up. I am not asking you to take that on faith. What I would like to do is to invite you to look at the last 10 years again, identify underlying trends, and play them forward. In the last 10 years, almost to the day since we launched, Lemonade has spent about $500 million cumulatively on technology. Half a billion dollars spread over 10 years. During that same decade, the large incumbents, insurance companies, have reported spending between $1 billion and $4 billion each year.
That is $10 billion-$40 billion each over the same decade during which we spent half a billion. Let that detonate in your mind for a second. Large incumbents each outspend us on technology alone by 1.5 to 2 orders of magnitude. Taken together, the incumbency has spent about $2 trillion on technology, outspending us 4,000 to 1, several orders of magnitude. So yes, their spending eclipses ours. But what you saw here this morning, our technology, absolutely eclipses theirs. That tells you something profound, something structural. If our advantage could have been eroded by throwing money at the problem, it would long since have vanished. But it has not eroded, it has compounded. The gap between us and the incumbency has never been greater. Our systems and theirs are more incomparable today than they were two years ago, let alone four, six, eight, or 10 years ago.
None of this means that incumbents are sitting on their hands. Absolutely not. They are led by smart people who are investing heavily and doing what they can to drag their companies, kicking and screaming if they have to, out of the 19th century and into the 21st. Bless you. Here is the final point. The frontier is moving faster than they are, and so are we, which means that for all of the energy, for all of the investment, they are actually falling further behind with every passing day. The gap is not closing, the gap is gaping. That is why our technology, its advantage have become more pronounced with every passing year. It has now passed key thresholds, and the wins are stacking up, they are now manifesting across our P&L.
Roger Federer once said that he won close to 80% of the singles games he played in tennis, but only 54% of the individual points. 80% of the games, only 54% of the individual points. I find that illuminating. One of the greatest careers in the history of tennis was built on an edge that point by point looks like a coin toss. My point is, small advantages can produce extraordinary outcomes when they operate again and again and again. That is why we have spent this morning showing you the individual decisions and how they are made and operated inside Lemonade, which customers to acquire, which risks to take, how to price them, how to handle a claim. Each decision perhaps matters little, but that advantage compounded across millions of decisions matters enormously. Changes in degree amount to a change of kind. Lemonade is now the low-risk, high-reward play.
With that, I want to take the opportunity to acknowledge the incredible humans behind the machine. A few of us were on stage this morning, but behind us stand a team that is smart, curious, energetic, and enthralled by the possibilities technology is unlocking, and I am ever grateful to them. I want to thank you, our investors and prospective investors. Thank you for your time, thank you for your attention, thank you for your trust. We are now cash flow positive, and we expect to be EBITDA positive this quarter and thereon out. Over the coming years, we see our path clear to continuing the compounding that brought us here, compounding our data, compounding our intelligence, compounding our customer count and our revenue, compounding our profits and our profitability, and yes, compounding our investors' returns. That is what makes Lemonade the low-risk, high-reward play.
I hope we have convinced you of the incredible future we believe in. At a bare minimum, I hope I have convinced you that we believe in it. Thank you so much.