Earnix CEO Cuts Insurance Pricing Lag From Months to Days

London, 14 October 2026 — EBM Newsdesk Analysis — By Nick Staunton

Robin Gilthorpe, chief executive of Earnix, runs an AI company with more than $100 million in annual revenue. It has a customer list that includes Aviva, Generali and Hiscox, and a backer, Erel Margalit of Jerusalem Venture Partners, who has already paid early investors 8.7 times their money. Yet when I sat down with the two of them, the most striking thing Gilthorpe said had nothing to do with models or agents. It was about time: insurers routinely take months to put a risk they can already see into the price of a policy.

That lag is the real story of AI in insurance. Generative AI took over the consumer internet because nobody needed permission to use it. Regulated industries do. In insurance, banking and healthcare, a clever answer that breaks the rules is worse than no answer at all. Earnix is betting that the next wave of AI value will go to specialists who already know the rules, not to the giants who wrote the chatbots.

Two Kinds of AI

Margalit frames it as wild agents against vertical ones. A general-purpose AI can be brilliant and still dangerous in the wrong setting. He compares it to drug discovery: anyone can have an idea for a new compound, but someone has to check it won’t poison the patient.

His examples come from compliance failures rather than science fiction. Banks such as TD and HSBC have paid billions in fines for letting money laundering slip through systems that looked perfectly respectable. “These are bad things that look kosher,” he said. The transactions passed because nobody was approving them the way a regulator would.

AI agents raise the stakes. Every time an AI project runs through a company’s systems, it touches the database. If neither the user nor the agent knows the rules, the damage scales with the speed.

Gilthorpe puts the commercial case more plainly. “In these regulated markets in particular, deep mastery wins,” he said. Insurance grew up over 600 years. A practice standard in the London market can be alien in global specialty lines and different again in personal lines. The horizontal AI firms will compete fiercely with each other. But regulated verticals, in his view, sit “well down their list” because they demand so much context.

Fifty-One Rulebooks

Earnix operates in 36 countries, and the United States alone is 51 jurisdictions. The restrictions differ from state to state, and so do the formats for filing a rate change. Gilthorpe calls automating that “kind of a big deal,” and he is right.

The trick is leverage. Compliance work is no competitive advantage on its own, because everyone has to do it. But Earnix does the work once per jurisdiction and reuses it for every customer there.

Twenty years of predictive AI in pricing and underwriting forced the company to build a governance framework early. Gilthorpe says the same framework now covers its generative AI and the newer agentic tools. Every query an agent runs on the platform stays inside the same rules of the game.

From Months to Days

This is where the argument stops being abstract. Between an insurer spotting that a risk has changed and a customer being able to buy a policy that reflects it, the gap is typically months.

“If you’re trying to trade on the stock market and you have to wait for months, you’d never make any money,” Gilthorpe said. “It would be gambling. It wouldn’t be trading.”

Earnix says it has cut that cycle from months to days. Inside the system itself, it takes minutes. One Canadian customer started from a five-to-six-month delay, and Gilthorpe promised the full before-and-after figures would follow.

Speed matters because insurance runs on capital. An insurer cannot write a policy unless it holds the capital to back the potential loss. That capital comes from equity, debt or reinsurance. If an insurer cannot see its risk quickly, it has to hold too much capital as a buffer, and spare capital earns nothing.

The alternative is worse. “When they say we have a 108 per cent combined ratio, that means in the long term they’re losing their shirts,” Gilthorpe said. “Ultimately they can’t pay claims.”

He is clear that the gain is not about cutting staff. It comes from Earnix’s machine learning reaching a more precise answer under awkward conditions, and then pushing that answer into the market. Most insurers’ technology stacks were never built for their parts to talk to each other. “We don’t just help you do the analysis better,” he said. “We help you come to a decision and make it happen.”

Stop Counting Tokens

I put to him the figure doing the rounds: 81 per cent of insurers say they have adopted AI, yet few can show it has paid. Is that bad design?

He thinks it is mostly bad measurement. Some companies have set staff targets for how many AI tokens they use. That is fine as a nudge in the first months, he said, but it soon stops meaning anything. Willingness to experiment matters, and so does going back when something fails the first time. “But if you’re not focused on business impact and actual business outcomes,” he said, “then you’re doing it wrong.”

Earnix tries to practise this in its own pricing. It does not charge per seat. Its fees track the size of a customer’s premium book and the volume of transactions, which is how insurers earn their money. If the customer grows, Earnix grows. It is a quietly radical choice in a software industry that still loves the per-user licence.

Why Not Just Raise Again?

The money behind Earnix is unusual too. Last year JVP closed a $290 million continuation vehicle led by TPG, one of the largest single-asset deals of its kind by a venture firm.

Gilthorpe explains the logic simply. JVP had backed Earnix from two funds of different vintages. Investors in the older fund had signed up for roughly seven years and were ready for their money back. The vehicle let them cash out at 8.7 times without forcing a sale of the company. “JVP and the company get to focus on completing the story,” he said, without being distracted by investors who want an exit.

Margalit sees the same pattern he rode at CyberArk, which grew into a business worth around $25 billion: an old industry rebuilt on new technology. He has said Earnix needs about $300 million in revenue before a stock market listing makes sense.

The Quant in the Chair

Gilthorpe has been chief executive for just over four years. He came up as a quant. He passed the US securities exams rather than the actuarial ones, then spent about three decades as a practitioner in capital markets and insurance. Later he led technology businesses serving those industries. He had already been a CEO before taking this job.

That background shows. He talks about combined ratios and capital charges the way most software executives talk about user growth.

My Read

The AI conversation in Europe is still dominated by the model builders and their valuations. Gilthorpe’s case is less glamorous and more convincing. In regulated industries, the winners will be the firms that already know where the tripwires are and can turn faster decisions into hard returns. Shortening a five-month pricing lag is not a demo. It changes how much capital an insurer has to hold, and therefore whether it can pay its claims.

The test is the next two years. Earnix has to get from $100 million to $300 million in revenue and prove that its agents behave as well as its spreadsheets. If it does, the most important AI company in your insurance policy will be one you have never heard of.