Lemonade Insurance: Did AI Disrupt Insurance or Just Hype?

Introduction

In the summer of 2020, an insurance company called Lemonade went public at a $1.6 billion valuation. On the very first day, the share price doubled, bringing the market cap to over $3 billion. Using a digital-first platform powered by artificial intelligence, Lemonade promised to disrupt the $7 trillion insurance industry.

At first, this seemed like a great idea. The insurance industry is dominated by massive companies, some of which were founded over 100 years ago, and their business practices haven’t changed a whole lot since then. This has made the industry ripe for disruption.

With Lemonade’s AI-based algorithms, they promised to process claims far more quickly and efficiently. This allowed them to offer very low premiums to consumers. Unlike traditional insurance companies, which seem to do everything to make it as difficult as possible to make a claim, Lemonade is not motivated to withhold claims. They make a flat fee, and excess premiums are donated to charity.

Lemonade used their IPO proceeds to expand rapidly, both to new geographical markets and coverage areas. But despite the hype, the company has still never made a profit, and its share price has declined by more than three-quarters since the IPO.

In this write-up, we’ll take a deep dive into Lemonade and why disrupting the insurance industry may be quite a bit harder than many technologists had hoped.

Insurance Industry

To understand how Lemonade differentiates itself, we first need to understand how the insurance industry works. Let’s say you just bought a house and want to get insurance on it. For many people, their home can be the majority of their net worth.

If you have a mortgage, its value is likely greater than your net worth. If your house burns down, you’d be stuck paying mortgage payments for the next couple of decades for a house you can’t even live in. This could be financially catastrophic.

Thus, we can see why insurance is a $7 trillion industry. A lot of people are willing to pay a couple thousand per year for the peace of mind of knowing they’ll be protected from catastrophe. Let’s say you just bought a new house and want to get insured. How do you go about doing this?

Typically, you go to a local insurance office and talk to a licensed agent. The agent may be an independent contractor or an employee of the company. Either way, they almost always work on a commission. They’ll explain to you the various coverage options, how much you can expect to pay, etc.

Insurance companies make a profit on the majority, if not the vast majority, of their customers, but they have huge losses on a small number of their customers. For example, if you pay $1,000 per year for homeowners insurance on a $200,000 house and a tornado completely destroys your house, the insurance company will obviously make a massive loss on you.

From the perspective of the insurance company, it’s all about statistics. For each individual, they have to judge the likelihood of making a claim and charge a high enough premium that the expected value is positive. If your home is too risky, they may decline to offer you coverage at all.

To make this determination, the insurance company needs a lot of information. You’ll need to fill out a large amount of paperwork, giving details about yourself, your house, etc. In some cases, the insurance company will send someone to inspect your home. They’ll estimate the replacement cost and chances of things breaking down. The insurance agent will input all the information into software created by the insurance company. This software will decide your premium.

The premium depends on a wide variety of factors. For example, if your house is very old, you will likely pay a higher rate as it’s more likely to need significant maintenance in the near future. If you live in an area prone to tornadoes or other extreme weather, your premium will also be significantly higher.

Once you have an insurance policy in place, let’s say a storm rips off your house’s roof, so you file a claim. This often also requires significant paperwork. The insurance company will often send an adjuster to visit your house to survey the damage and determine how much they will pay to repair the damage.

Many people have an unfavorable view of the insurance industry for a number of reasons. Firstly, both signing up for a policy and filing a claim can be time-consuming processes, which require copious amounts of paperwork. Secondly, because most insurance agents work on a commission, they have an incentive to sell you the most expensive policy, as this will give them the biggest commission.

This leads many people to feel like they’re being pressured into getting more insurance than they need. Finally, insurance companies lose money every time you make a claim. They’re thus incentivized to minimize the number of claims they pay out. Many policyholders feel that their insurance company makes it unnecessarily difficult to file a claim.

Lemonade

Now that we understand the basics of insurance, we can turn our attention to Lemonade. Lemonade was founded in 2015 by Daniel Schreiber and Shai Wininger. Schreiber formerly served as a president of Powermat Technologies, a company that makes wireless cell phone chargers.

Wininger was a co-founder of the online freelance marketplace Fiverr. Both men were technologists with no prior experience in insurance. Far from making them unqualified, they believed their lack of experience was an advantage. Traditional insurance companies were too old-school and lacked innovation. They could reinvent insurance from the ground up with a tech-first mindset.

They garnered the attention of the Japanese venture capital giant SoftBank, which invested $120 million into the company in 2017. This was just two years after the company was founded. Here’s a clip of Lemonade’s CEO explaining their business model in 2018:

“When you buy insurance, you’re talking to a chatbot. Just download the app or go to the website. If you’ve spent more than 90 seconds buying insurance, you’re doing something wrong. So it’s just a friendly chat that you go through, answer a couple of questions, and you’re done. All of the underwriting, all of the binding of the insurance is done like that. No human does a human being internally ever review what the bot has decided.

Well, where the humans come in is where claims get paid. We have bots that pay claims as well. A third of our claims are paid algorithmically within three seconds of submission. So literally, you’re making a claim, you chat to our chatbot—it’s called AI Jim—you tell AI Jim, ‘Hey, I got my laptop stolen, this, that, the other,’ and in a third of the cases, the money is in your account within three seconds.

You don’t have to show any proof?

Well, yeah, depending on the nature of the claim, it will ask you this stuff.

So I don’t want to suggest that some of your customers game your AI here, but what kind of claim would the AI bot accept in 30 seconds?

Well, in homeowners and renters insurance, there’s a lot of what’s known as high-frequency, low-severity claims: your laptop was stolen, your bike, your camera, that kind of stuff. The AI will pay those things all day long and within three seconds. If the house was burnt to the ground or you’re being sued by your neighbor or stuff like that, that’s obviously more protracted.

We take a flat fee, and if there’s money left over at the end of the year, it goes to a charity of your choosing rather than being pocketed by us. That neutralizes some of the conflict of interest. We remind you, ‘Hey, please remember that unclaimed money is going to the local soup kitchen that you designated as your give-back cause.’ You might moderate your behavior and embellish your claims less.

So people behave better?

That’s right. And if we have no interest in denying your claim because we’re not going to pocket that money anyway, we may behave differently. So this is also about trying to create a system that works better and brings out the best in humanity, whereas insurance today frankly oftentimes brings out the worst in humanity.”

Lemonade’s Digital Transformation and Risk Management Strategy

The Lemonade sign-up process is completely digital. You download an app and talk to a chatbot called AI Maya. You input your home address, and it asks you a bunch of questions, such as when was the last time your roof was replaced, how many people live in your house, etc.

Within a few minutes, it’ll give you a quote. Lemonade claims to have advanced AI models to determine your riskiness. This is much faster than all the paperwork you would need to complete with a traditional insurance agent.

When you need to file a claim, they have another chatbot called AI Jim. You tell the chatbot about your claim, then take a short video of yourself explaining what happened. Lemonade uses facial recognition to verify your identity. You may be asked to provide other evidence.

For example, if your house was burglarized, they may ask for a police report. For small-dollar claims like renters insurance, the processes are completely automated with AI, and you can often get your claim paid within minutes. For larger claims, Lemonade will send an adjuster to your home to survey the damage, just like any other insurance company would.

Finally, Lemonade claims it does not make more profit if it denies customers’ claims. But how is this even possible? When an insurance company pays out a claim, the money has to come from somewhere. CEO Daniel Schreiber gave an explanation in a blog post. Lemonade makes a flat fee on each insurance policy it sells. If the claims are less than expected, the remainder is given to a charity of the customer’s choosing. If Lemonade denies a claim, this just means more money will be donated to charity; it won’t increase Lemonade’s profits.

While this sounds great, it also leads to an obvious question: what if the claims are greater than expected? Lemonade can’t claw back money from the charities, so how does it work? To understand how Lemonade accomplishes this, we need to understand reinsurance. Reinsurance is essentially insurance for insurance companies. Let’s say you pay your insurance company $1,000 per year. The insurance company might pay 10% of that to a reinsurance company. If the insurance company receives an unusually large number of claims, they can make their own claim to the reinsurance company.

Hypothetically, let’s say the reinsurance company will pay for any claims in excess of $5,000. Let’s say, in a given year, the end customer doesn’t make any claims. This is the best-case scenario: the insurance company makes $900 of profit (this is your $1,000 premium minus the $100 they pay for reinsurance). The reinsurance company makes $100 of profit.

Now let’s suppose a window breaks in your house and you make a $2,000 claim. The insurance company makes a $1,100 loss, and the reinsurance company still makes a $100 profit, as it’s still below the reinsurance threshold. Finally, let’s say a windstorm blows off your roof, and it costs $30,000 to replace.

The insurance company’s loss is capped at $4,100. This is the $5,000 cap minus the $1,000 in premiums you paid that year, plus the $100 they paid to the reinsurance company. The reinsurance company makes a loss of $24,900.

We use the example of a single end customer for simplicity. In reality, reinsurance plans are done on an aggregate scale. The insurance company has enough reserves to handle the amount of claims they would expect in a normal year. But if there’s a large natural disaster that impacts thousands of customers, that’s when the reinsurance kicks in. While reinsurance caps the insurance company’s losses, it’s only activated in extreme circumstances. Most of the time, the customer claims are paid from the insurance company’s own reserves.

How Lemonade’s Digital Process and Reinsurance Model Work Together

The Lemonade sign-up process is completely digital. You download an app and talk to a chatbot called AI Maya. You input your home address, and it asks you a bunch of questions, such as when was the last time your roof was replaced, how many people live in your house, etc. Within a few minutes, it’ll give you a quote. Lemonade claims to have advanced AI models to determine your riskiness. This is much faster than all the paperwork you would need to complete with a traditional insurance agent.

When you need to file a claim, they have another chatbot called AI Jim. You tell the chatbot about your claim, then take a short video of yourself explaining what happened. Lemonade uses facial recognition to verify your identity. You may be asked to provide other evidence.

For example, if your house was burglarized, they may ask for a police report. For small-dollar claims like renters insurance, the processes are completely automated with AI, and you can often get your claim paid within minutes. For larger claims, Lemonade will send an adjuster to your home to survey the damage, just like any other insurance company would.

Finally, Lemonade claims it does not make more profit if it denies customers’ claims. But how is this even possible? When an insurance company pays out a claim, the money has to come from somewhere. CEO Daniel Schreiber gave an explanation in a blog post. Lemonade makes a flat fee on each insurance policy it sells. If the claims are less than expected, the remainder is given to a charity of the customer’s choosing. If Lemonade denies a claim, this just means more money will be donated to charity; it won’t increase Lemonade’s profits.

While this sounds great, it also leads to an obvious question: what if the claims are greater than expected? Lemonade can’t claw back money from the charities, so how does it work? To understand how Lemonade accomplishes this, we need to understand reinsurance. Reinsurance is essentially insurance for insurance companies.

Let’s say you pay your insurance company $1,000 per year. The insurance company might pay 10% of that to a reinsurance company. If the insurance company receives an unusually large number of claims, they can make their own claim to the reinsurance company.

Hypothetically, let’s say the reinsurance company will pay for any claims in excess of $5,000. Let’s say, in a given year, the end customer doesn’t make any claims. This is the best-case scenario: the insurance company makes $900 of profit (this is your $1,000 premium minus the $100 they pay for reinsurance). The reinsurance company makes $100 of profit.

Now let’s suppose a window breaks in your house and you make a $2,000 claim. The insurance company makes a $1,100 loss, and the reinsurance company still makes a $100 profit, as it’s still below the reinsurance threshold. Finally, let’s say a windstorm blows off your roof, and it costs $30,000 to replace.

The insurance company’s loss is capped at $4,100. This is the $5,000 cap minus the $1,000 in premiums you paid that year, plus the $100 they paid to the reinsurance company. The reinsurance company makes a loss of $24,900.

We use the example of a single end customer for simplicity. In reality, reinsurance plans are done on an aggregate scale. The insurance company has enough reserves to handle the amount of claims they would expect in a normal year.

But if there’s a large natural disaster that impacts thousands of customers, that’s when the reinsurance kicks in. While reinsurance caps the insurance company’s losses, it’s only activated in extreme circumstances. Most of the time, the customer claims are paid from the insurance company’s own reserves.

Challenges and Criticisms

Now that they’re taking more risk onto their own balance sheet, their claim that they don’t benefit by denying claims is becoming less and less true. One of the most important metrics in the insurance industry is the net loss ratio. This is the amount of money the insurance company pays out in claims as a percentage of premiums earned minus reinsurance expense.

In 2023, Lemonade’s net loss ratio was 89%. That means that Lemonade pays out 89 cents in claims for each dollar of premiums it collects. The five largest publicly traded property and casualty insurance companies in the US are Progressive, Allstate, Travelers, Chubb, and AIG. Their net loss ratios range from 61% to 81%. Lemonade’s loss ratio is far higher.

So what happened to Lemonade’s advanced AI algorithms? If they really had a data advantage over their competitors, you would expect to see them have a lower loss ratio, not a higher one. If you look at the questions their AI chatbot asks prospective customers, it’s all pretty basic stuff like your address, when’s the last time your roof was replaced, and do you have fire alarms installed. These are all basic questions that traditional insurance companies have been asking for decades. Lemonade claims that they collect 100 times more data than their peers, but I really can’t see where this supposed data is coming from.

Lemonade’s founders were technologists with no prior experience in insurance. They thought they had an advantage because they had the tech mindset to disrupt this otherwise sleepy industry. In this, they were mistaken. Just because the incumbent insurance companies have been around for 100 years doesn’t mean they’re technologically incompetent.

If you go to the careers pages of any of the large insurance companies, they’re always hiring data science PhDs with cutting-edge technical knowledge. The established companies also have many decades of historical claims data which they can perform simulations on. Lemonade may look more tech-savvy due to their hip marketing materials, but as far as data science is concerned, they’re not really doing anything new.

To be fair, Lemonade has tried to incorporate AI into various parts of the process, but these efforts have yielded mixed results. For example, when you submit a claim through the Lemonade app, you need to record a short video of yourself explaining what the claim is for. They initially did this as a way to verify the customer’s identity. In a desperate attempt to convince investors that their AI is doing something, Lemonade published a tweet explaining more details. When the customer uploads a video of themselves, the AI picks up on non-verbal cues to determine whether or not the claim is fraudulent.

Having an AI process claims based on a customer’s video is obviously a horrible idea. The AI could easily start profiling people based on gender, race, etc. Lemonade got so much backlash that they quickly deleted that tweet and posted a clarification. They said they do not use physical or personal features to deny claims.

But if this is the case, why even require customers to upload videos? They now say it’s because humans are less prone to lying when looking at themselves in a mirror or selfie camera, so it’s purely psychological and has nothing to do with AI. In my opinion, they’re probably telling the truth here.

If they had really developed an AI so advanced that it could tell if someone is lying just from a selfie video, why would their loss ratio be among the worst in the industry? In the original tweet, they’re probably just embellishing their AI as they thought this would look good to investors. I doubt they even had the capabilities to do what they claimed.

Over the past decade, institutional investors have allocated hundreds of billions of dollars into what they thought were disruptive tech startups. They believed that all industries could be revolutionized with the internet and artificial intelligence.

When Lemonade showed up with a vision to disrupt the $7 trillion insurance industry, this was too attractive of a proposition to ignore. The company’s disastrous performance should serve as a cautionary tale. Unlike what many tech enthusiasts think, AI is not a magic pill that can solve all the world’s problems.

Alright guys, that wraps it up for this video. What do you think about Lemonade? Let us know in the comments section below.