Inside almost every insurance company’s list of customers, some policies are quietly losing money and sometimes nobody knows which ones. McKinsey says this is true for 20% to 30% of a typical insurer’s policies, even at companies that look profitable overall.
Soteris, a machine learning company backed by Y Combinator, is looking directly at this problem and today the company launched a product built to find those losing policies. No price changes, no new customers needed, no new hires, they do a better look at the business the insurer already has.
“Every insurer knows they’re writing policies that will lose them money. They just can’t find those policies with the resources currently at their disposal,” said Sunit Shah, founder and CEO of Soteris.
“That’s the blind spot we built Soteris to close. For the first time, an insurer can look at a single policy and know exactly what it’s worth, in time to act on that information.”
The company has run several proofs of concept for insurers, where it has seen book EBITDA increases of between 70% and 125%, which means an insurer could more than double its bottom line by using Soteris’s new product.
Six years of saying nothing
The company was part of the Y Combinator 2019 cohort, is based in Richmond, Virginia, and quietly began selling its first product to insurance companies in 2020, after which it said nothing publicly for six years.
During that time, without any advertising, word of mouth alone got the company’s software used to evaluate more than 100 million insurance policies, worth over $180 billion USD combined.
The silence matched how insurance companies actually buy software: big insurers take months, sometimes over a year, to vet a new vendor, check security, compliance, and get budget approval.
Soteris says that once a company decides to use its product, it only takes about three months to get it running. The slow part is everything before that decision.
Launching with no track record would have meant asking these cautious buyers to take a leap of faith. So instead, Soteris spent years quietly building a list of real customers first, and is only now pushing more into the public eye.
The money followed the same pattern. The company is now confirming over $8 million USD in early funding, led by Spider Capital, with participation from Intact Private Capital, Amplify Partners, Foundation Capital, the Webb Investment Network, and Overlook Ventures.
Why the average number hides the problem
In insurance, a company sells a policy today but doesn’t find out what it actually cost, them, how much they had to pay out in claims, until months later – if ever. Looking at any single policy, therefore, tells you almost nothing about whether it was priced correctly.
The old way around this hurdle was to group similar policies together; say, all homeowners in a certain zip code with similar homes, and look at the group’s average outcome instead of any one policy. Grouping enough similar policies together turns a confusing, unpredictable one-by-one picture into a reliable pattern. But this only works if the group is big enough.
And here’s the catch: inside any one of these groups, some individual policies are losing money, others are making money, and they just average out, meaning losers were never spotted. Insurers have gotten increasingly better at slicing these groups more finely via more advanced tools, but there’s a hard limit to how small a group can get before the averaging stops being reliable – and that limit hasn’t moved in decades.
Soteris says it found a way around that limit. Instead of sorting each policy into one group, its software builds an enormous number of different possible groupings of the same customers, looking at them from every possible angle at once, figuring out where each individual policy lands across all of those groupings combined.
A human working with spreadsheets might come up with a few dozen useful ways to group policies. Soteris says its software can generate millions, or even billions, of these groupings, depending on how much information it has to work with. The result, the company says, is that it can judge each policy almost on its own in under a quarter of a second. In fact, insurers using their product saw claims cost improve by 5% to 15% within a year.
A profitable-looking policy can still be a bad deal
While rolling that first product out to real customers, Soteris ran into a separate problem that had nothing to do with predicting claims. Throughout specialty insurance deals in the U.S., three different companies share one policy: one sells it to the customer, one lends its license to make the policy legal, and one actually puts up the money to cover potential payouts.
Each of those three companies gets a different cut of the deal, so a policy can look completely fine on paper with low claims and healthy pricing, and still be a losing deal for whichever of those three companies got the short end of the contract terms.
This three-way split has become much more common across specialty insurance over the past ten years. Soteris built its new product specifically to catch that second, harder-to-see kind of loss.
The original product reports its results the way an actuary, an insurance company’s in-house numbers expert, would: in terms of claims costs. The new product, however, reports results the way a CEO or a board member would: in terms of actual profit.
Under the hood, it’s a similar kind of model. It’s just answering a different question, for a different person.
An investor at Spider Capital, Minsoo Chi, described this less as a pivot and more as the next logical step, as Soteris had already moved from judging insurance policies in large groups to judging them individually.
“With his rare combination of finance, insurance, and quantitative experience, Dr. Shah is uniquely positioned to crack the toughest mathematical problems in insurance,” Chi said.
Featured image: Vlad Deep via Unsplash+
Disclosure: This article mentions a client of an Espacio portfolio company.
19 c