Parametric MGA Descartes on Gen AI, expansion and distribution

A recent episode of the Voice of Insurance podcast saw host Mark Geoghegan joined by Tanguy Touffut, the CEO and co-founder of Descartes Underwriting, a pioneering parametric insurance managing general agent (MGA).

It’s been quite a growth story for Descartes, quadrupling in size from $50m gross written premium two years ago – when the firm last appeared on the podcast – to a projected $200m by the end of 2024.

Parametric insurance makes payments based on predetermined parameters, such as weather or seismic activity; it does not reimburse policyholders based on a judgment of their losses. This means faster and easier processing of claims, and price certainty for the consumer.

Touffut is interested in fast-moving developments for using generative artificial intelligence (Gen AI). He said he expects applications to appear “in the next two to three years”. However, insurers and buyers need to understand the basics first, he suggested, adding that this means putting it in the hands of senior, experienced employees first.

“It’s a compression of lots of data, lots of knowledge, across the world. Start with the basic knowledge first, and be able to take a step back and analyse the data – otherwise more data won’t make a difference,” he said.

Mark asked the familiar ‘build or buy’ question about using proprietary technology or buying off the shelf applications.

 “We want to have all our models in house, something that is crucial,” Touffut declared.

“One of our key convictions is that, if you’re an insurer or an MGA, your model should be in-house. There should be no black box. You need to feel comfortable with both your data and your algorithms to make sure that you can offer a fair price to brokers and clients,” he added.

For Gen AI this is not the case, with “billions and billions” being invested by technology firms, an incremental or layered use is preferable, he suggested, rather than a heavy investment from scratch. “It’s more tool that will be on top of in-house tools, to make them better,” he said.

Parametric products are much more efficient than the traditional indemnity form of insurance, Touffut argued, using the example of a wildfire claim.

“Through satellite imagery, you’re able to detect if it’s burnt or not burnt. If you send loss adjusters on the ground, even if you send 50 of them, the ability to capture the true loss through computer vision and algorithms, is much better than through the work of experts,” he said.

He said he sees parametric as “a means, not an end”, which means if satellite data or other indices replace surveyors on the ground, the product is arguably parametric.

Parametric works for some perils and lines of business more than others, Touffut acknowledged.

“Is not true for all products and for all lines of business, but in some cases, we’re in a situation where we able to beat threshold covers, in that case, replay them with a lower business risk and a fair price, because we will be able to have no buffer,” he said.

The buffer refers to uncertainty, leading to higher price. “If you can get rid of the insanity, then you can get rid of the buffer,” he added.

Mark noted the tricky example of hotel occupancies, a previous parametric focus, and one that something like the pandemic caught off-guard, much as it did many other traditional insurance policies, leading to disputes and unhappy policyholders.

Touffut admitted the same issues of what is covered and what’s not apply to parametric claims, with the same necessary focus on clarity of wordings, despite the objectivity afforded by a data-based trigger.

“Sometimes there could be a scenario leading to a scenario, leading to a scenario something that is not at all in the spirit of the policy,” he said. “We cover flood. If you want a quake policy, you have to go for a quake policy. It’s the same if you want cyber cover, it means potentially we won’t cover a fire in your factory. We have to be clear.”

The hard market for property catastrophe reinsurance has led to higher attachment points for reinsurance, creating a window for parametric products to snap up lower layers – provided the risks, pricing and data are accurate and acceptable.

“We don’t want to hide behind exclusions and we don’t want to use a buffer. It means we have to be much better in the way we understand the risk,” Touffut said.

This means a focus on hiring software engineers, risk modellers and physicists. The MGA’s headcount has surpassed 200 people, building a team with the right expertise in meteorology, seismology, and data science, for example.

However, some of this increased headcount and footprint is about the basics of insurance distribution, rather than advanced science or mathematics. The MGA’s focus on technology does not dispel the need for a global footprint, he explained, with a focus on “being close to brokers and clients”.

This is particularly true for the US market, he emphasised. Descartes has expanded to eight offices in the US, four in Europe and four in Asia Pacific (APAC). Next on his agenda is opening a Mexican office, to cover that country and other Latin American markets currently being served by the MGA’s office in Madrid.