Black box AI is hiding new risks around data confidence – Cactus’ Robinson

There is a widespread suspicion that artificial intelligence (AI) and its insurance industry technology exponents are more-or-less out to replace the underwriter.

Instead, James Robinson (pictured), the founder of Cactus, says his company’s flagship platform, Cactus Risk Studio, is designed to help practitioners organise, interrogate and trust the information they already work with every day.

“We focus on helping organise and use risk and financial information, with a focus on making sure that they can trust and rely on whatever information they get out of the sources they use,” Robinson says.

“They often can’t by jumping straight to a AI black box model,” he adds.

Bermuda-based Cactus works with brokers and carriers, but Robinson says the real day-to-day users are underwriters with signing authority.

“They do about half a billion dollars a year of premium at the moment, processed through Cactus Risk Studio,” he says.

He sees a problematic tendency across the market to treat data handling as a junior level task, when, in reality, it often requires more senior judgement.

Inexperienced data handlers are unlikely to catch an error in a source document, he suggests.

“If you take the approach that what we might call mundane data entry, it’s an activity for juniors, that’s fine, but the thing that they can never do, that seniors can, is they can say you’ve sent me the wrong information,” he explains.

Robinson provides a simple example.

“You could read that slip perfectly, but apart from one character, you could miss a zero,” he says.

“You miss a zero, the amount of premium you’ve booked in systems out by a factor of 10… you put a decimal point in the wrong place, that could be four pixels.”

For Robinson, this is where the current enthusiasm around AI can generate hidden risks.

“The moment that a black box AI system touches your process, your trust can drop to zero,” he warns.

“When the error is then something that very plainly only an AI would make, you’re in a mess. It’s a reputational hit as well as anything else,” he adds.

He stresses that Cactus is not anti-AI, but says that the technology must sit within a framework that preserves auditability and confidence in outputs.

“There’s a huge opportunity there,” he says. “The problem that exists today is [when using AI] how do you make sure you can have confidence in the information you have.”

Timing also matters, particularly given the recent reinsurance renewals.

Around peak renewal periods such as 1 January, Robinson argues that access to clean, reliable information can translate directly into commercial advantage.

“If someone’s trying to fill a tower and they’re missing a layer and it’s just about to bind, you might get a better rate than you would have done if you filled it at least a month earlier,” he says.

“Having that ability to be nimble is crucial, and often that’s about having good systems and information, which lets you focus on decisions.”

Cactus has clients in Bermuda, London and the US, and Robinson says the firm’s participation in the Lloyd’s Lab has helped build relationships in the London market.

A recent merger with Dubai-focused property reinsurance tooling firm Cat X is also expanding its geographic and product footprint.

But the core message remains consistent, he stresses.

“Good information lets you focus on decisions, rather than having to say, okay, I’ve got six days lead time,” Robinson says. “That’s where there can be big advantages.”