Lemonade

What it really costs an insurer to handle one claim

When people talk about the price of a policy, it is easy to assume it is set mainly by risk. In practice there is a second component, large and hidden: how much it costs the company to run itself. Acquiring a customer, underwriting, and human handling of every claim all add up in every policy, whether or not anything is paid out in the end.

A claim that goes through a clerk costs money even when it is completely simple. Worse, it takes time, and the moment of a claim is exactly the moment a customer decides whether their insurance is worth anything. Slowness here hurts twice: in cost and in trust. Older companies tried to solve this by adding automation on top of an existing call center. That lowers costs at the margins but does not change the structure.

This is the angle that makes Lemonade interesting for an AI encyclopedia. It did not add automation to an existing human process. It built the process to be automated from the start, and left people only for the exceptions. The bet is that a different cost structure, and not impressive technology in itself, is what eventually makes it possible to sell for less.

Three seconds: the 2016 record and what happened in it

On 23 December 2016, Lemonade closed a claim in three seconds. According to the company, between 5:49:07 and 5:49:10 its bot, AI Jim, reviewed the claim, cross-checked it against the customer's policy, ran 18 anti-fraud algorithms on it, approved it, sent the bank a wire instruction for $729, and notified the customer.

The company itself noted that Guinness World Records could not list this as an official record, simply because there is no international standard for measuring how fast claims are paid. The interesting part is not the record but the list: step by step, all of it happened without a person. An important caveat: this was a small, simple claim, and complex claims still go to human handling today.

What matters in this list is that fraud detection is not a separate, later step but part of the same sequence. When no clerk reviews each case on its own merits, a model that spots suspicious patterns is not an improvement to the business model. It is a precondition for running it.

$44 that became $14: the number, and what it does not measure

If one number is needed to show what automation did here, it appears in the company's shareholder letter for the end of 2025. The average cost of handling a pet insurance claim fell from $44 in 2021 to $34 in 2022, $25 in 2023, $19 in 2024 and $14 in 2025.

The caveat that follows is not a technical detail, and it changes what the number means. In the original, the chart is titled "Cost of Handling Pet Claims," and it sits inside the discussion of that product line only. It is not an average across all of the company's claims. What is more, this is the line with the highest claim frequency, a feature the company itself presents as an advantage, because it creates more customer touchpoints and more data to learn from.

The company describes this level as industry-leading, and credits the decline to the growing impact of AI on claims and operations. Even within these limits the number is useful for a non-technical reader, because it turns "artificial intelligence" into something you can track over time: not answer quality or an impression of smartness, but cost per unit of work. The broader lesson is bigger than the number itself. Before comparing a metric across years, check exactly what it is measured on.

1.2 million more customers, 6% fewer employees

The second metric the company publishes concerns the ratio between size and headcount. According to its report, since the third quarter of 2022 it has added 1.2 million customers, while its number of employees fell by 6%.

The same picture shows up in money. Over the same period, in-force premium grew from $609 million to about $1.24 billion, while quarterly operating expenses, excluding growth spending, rose only from $87 million to $101 million. In other words, the business doubled and its operating machinery barely did. "Expenses excluding growth spending" is a cut the company chooses for itself, and it leaves out the marketing budget that brings in the customers in the first place.

The company describes the mechanism as a loop. Faster growth produces more data, data improves the segmentation and pricing models, better models improve the economics of each policy, and the resulting gross profit is reinvested in growth. This is the company's central claim about itself, and it is best read as a claim being tested over time, not as a fact.

Insurance for a car that drives itself

On 21 January 2026, Lemonade launched a product called Lemonade Autonomous Car, insurance built specifically for autonomous driving. It starts with Tesla cars that have FSD (Supervised), a system that still requires an alert driver, not fully autonomous driving. The per-mile rate on stretches where the autonomous system is engaged is about 50% lower than for human driving.

The technical mechanism is what makes this possible. With the customer's consent, the company uses the car's data interface to tell apart miles driven autonomously from miles driven by a person. It can even price by the software version installed in the car and by its sensors.

Shai Wininger, one of the company's founders, put the logic this way: a traditional insurer treats a Tesla like any other car, and AI like any other driver. The launch began in Arizona on 26 January, and in Oregon about a month later. The groundwork was laid earlier: mileage-based insurance and the acquisition of Metromile, which according to Lemonade brought with it data from more than 400 million trips collected by sensors installed in cars.

Where the company stands at the end of 2025

The fourth quarter of 2025 was the strongest in the company's history, by its own account. In-force premium reached $1.24 billion, up 31%. The number of customers stood at 2,984,513, up 23%, and the average annual premium per customer at $414.

Quarterly revenue grew 53% to $228 million, gross profit grew 73% to $111 million, and the gross loss ratio improved from 63% to 52%. There was still a loss: $22 million for the quarter. According to the company, pet insurance is now its largest product line by premium volume, with $439 million at the end of 2025. The company operates in the United States, the United Kingdom, Germany, the Netherlands and France.

Its targets for 2026 are in-force premium of about $1.63 billion and revenue of about $1.19 billion. It expects to reach positive adjusted EBITDA in the fourth quarter of 2026, and its first profitable full year in 2027. Until then it is a growing, loss-making company, a fact worth keeping next to any efficiency metric it presents.