The technology-insurance crossover – where generative AI fits in

Preventing the industry from falling foul of unintended AI consequences

The technology-insurance crossover – where generative AI fits in


By Mia Wallace

With three decades of supporting the insurance market in a variety of leadership roles at high-profile businesses to his name, Mind Foundry’s Selim Cavanagh (pictured) is no stranger to the “assortment” of reactions that accompany any period of operational or strategic transition.

Generative AI and its implications for the insurance proposition are no exception – and Cavanagh’s role as head of insurance at the AI solutions business is a natural evolution of his years of using connected data and technological innovations to help transform insurance businesses. Having served in senior roles at several high-profile companies including AXA, Lexis Nexis and ingenie, he was drawn to his present role by the mission of its founders – to create a future where humans and AI work together to solve the world’s most important problems.

“I’d got to thinking about how best to take my skill set around the technology-insurance crossover into other areas when I was approached by Mind Foundry – an extremely successful AI business spun out from Oxford University by two founders, who are still professors of AI there,” he said. “They’ve been researching and talking about the fundamental shifts in how we’re going to work in the future, what kind of jobs we’ll have and what impact that will have on society for some 30 years.

“They started Mind Foundry in 2016 and although they’d already had some success in the insurance sector without having specific people who knew the depths of the industry, they wanted to invest further for growth by hiring, and in 2023 I joined the business.”

Three pillars at the heart of the evolution of generative AI

Cavanagh noted that three fundamental pillars underpin the work being done by Mind Foundry, the first of which is transparency. The firm’s founders strongly believe that nobody is going to embrace AI or machine learning without first understanding how it works. An opaque ‘black box’ that nobody can see into does not engender trust or confidence, he said, not in society, not in regulators, not in consumers, and not in the industry as a whole.

“The second pillar is that of human-AI collaboration,” he said. “AI is only as good as it can be implemented and used by real people in what they do. It’s not about pushing a massive button, and then it all kind of happens and no-one understands how or why. If you do it that way, not only does it feel threatening, but it often has unintended consequences and doesn’t achieve the outcome you’re looking for.”

What’s important to remember at all times, he said, is that AI itself is not inherently clever. It doesn’t understand what the human wants to achieve and so it requires human collaboration in order to function effectively.

“The last pillar really is both a value and a technology called continuous metalearning,” he said. “This is essentially outlining that if the point of AI is to direct the machine to learn how best to find an insight and then deploy it, then effective AI should continue to do that inherently. It’s just a matter of doing it repeatedly, safely and reliably, which is why we’ve built out a platform that enables that continuous metalearning process.”

Where investment in AI currently stands in the insurance sector 

Looking across the insurance sector, Cavanagh said he can see quite substantial investment in the opportunities presented by AI over the last five-to-six years. However, he highlighted that this investment is still mainly in its ‘series of experiments’ phase, where tests of its capability are manually set up and run, powered by significant data sets used by insurers.

There are lots of clever and invested data scientists within these organisations looking to analyse, interpret, build and deploy models, he said, and they’re having some success in doing so. In reaction to this success, P&L owners in these areas are offering more investment to keep them doing what they’re doing. But essentially, it’s still mainly a series of preliminary experiments which are difficult to maintain, difficult to explain and difficult to prove that they comply with regulations.

“Having recently done some research around this, I’ve got an interesting stat that 50% of all time spent by data science teams is around the maintenance of the model, making sure that it’s explainable, fair and within the rules set by the regulators,” he said. “That’s making this a massively inefficient process.

“As use of AI in insurance grows and AI begins to penetrate more and more critical areas, you can see that becoming a bigger and more difficult problem to solve with so many resources eaten up by experts trying to maintain and manage these models. So, the focus of our platform is to try and take all these experiments and turn them into predictable, reliable, manageable, explainable and regulated pieces of insight that drive that business value without the associated cost and administration function sitting behind it.”

What’s next for generative AI in insurance?

The forecast from Mind Foundry is that there will be greater demand for this kind of process efficiency going forward. From conversations with P&L owners across the marketplace, Cavanagh said, it’s clear there is real appetite for where AI could go next but it’s being hampered by the significant operating expenses and risks entailed.

There’s also recognition of the massive opportunity cost represented here, he said, as time spent managing and maintaining these models is time businesses are not spending on rolling out new products and services, and learning more about their customers and their changing requirements. In addition, frequently, the model requires calibration because external market conditions have shifted and evolved which means businesses are required to sink additional capital expenditure into rebuilding their model.

“Our message to the industry is that AI is here to stay, if it’s managed properly,” he said. “If you get the explainability and regulations piece right, with the ability to maintain itself reliably and you can understand what’s going on across your whole enterprise, then the sky’s the limit in terms of using more AI to make yourself trade better, become more efficient, and respond better to customers’ needs. And that’s the journey we’re on at the moment.”

Related Stories

Keep up with the latest news and events

Join our mailing list, it’s free!