This week I read what just might be the last article you ever need to read about AI for insurance (except for this one, of course). It was written by a clever person named Ralitsa Nenkova, who’s Global Insurance Leader at Kyndryl, which says its mission is to “help customers in every industry ensure their essential systems will work when, where, and how they need them to.”  

The article, which was published on the website of the Association of British Insurers, makes five overall points (positioned as ‘strategies’) on the topic of AI in insurance technology. Seeing these, regular readers of Insurance Technology Diary will quickly understand why I liked the article so much.

Nenkova’s first point is something I have urged people to take to heart since long before I was your diarist. It says that you should begin by defining the outcome that you hope to achieve, rather than the technology you want to use. You don’t need an AI strategy, but your technology strategy should definitely incorporate AI.

It’s conceivable that AI may not be part of the solution to the pain point you wish to resolve (although not in the context of Nenkova’s article, since it is called The AI Native Insurer). However, when AI is involved – and perhaps even when it isn’t – one genuinely amazing possibility is to be bold with the outcome you target. Take the solution way beyond the pain point. Focus on a genuine process problem, but dream big. Even if the solution cannot be achieved today, it’s likely to be possible as soon as tomorrow, given the speed of technological advance.

The second strategy is another no-brainer. Concentrate first on applications which require low effort to implement but will deliver high impact. I once added a feature to a client’s underwriting workbench that meant no underwriter could exceed their authority without ringing a virtual alarm bell and causing the underwriting process to cease with a proverbial screeching of wheels and rising of smoke. Implementation was almost effortless, but the foregone hassle was gigantic. The client didn’t ask for it, but it’s one of the simpleton features users now praise more than any other.

Strategy 3, “Modernise your existing technology ecosystem with AI,” is really interesting. As Nenkova states, a big win can come from using Agentic AI to read existing code, then to create APIs to bridge old applications to a modern interface. The process should allow insurers to become agnostic about where their data sits.

More importantly, perhaps, though, such an approach would make data migration far less painful than even recent, AI-driven innovations allow. This would at last reveal the tunnel-ending light for companies reluctant to abandon legacy systems because they fear the pain of migration. It also supports another attractive goal I’ve suggested before: data translation on the fly, which would eliminate the need for the migraine-inducing process of implementing a different data standard for each and every market and trading partner.

Strategy 4 is about “transformational resilience” from the perspective of data breaches, and even “subtle interference” by third parties with systems, but it should also be about making sure you don’t let your virtual agents loose on an unsuspecting book of business. Controls are very important. Automation is great, but it needs to be checked and controlled, something else you’ve often heard me cry. The simple truth is that AI is fallible (for now, at least), and will sometimes chose a random, incomplete, or even erroneous outputs if that’s what it believes its users want.

Adoption, adaptation, and changing the ways employees work lies at the core of Strategy 5. I’ve touched on these themes many times before in this space, but many insurance-sector executives will run scared from Nenkova’s advice to: “Rethink your operating model to support AI-native ways of working.”

True, adoption of AI tools is key, and some processes will have to change (I realise I haven’t yet mentioned the ideal of eliminating bordereaux in favour of straight-through processing). However, to package the business change necessary to implement efficient, truly useful technology platforms as a new (let alone a ‘target’) “operating model” is bound to make insurance VPs seek comfort between their spreadsheets.

Instead, Strategy 5 must be about demonstrating to employees that they can be more efficient, which is to their benefit. In line with Strategy 2, aim first for the improvements that will have the highest impact on the reduction of employees’ mundane tasks, and you will win the hearts and minds from the first demo. Better yet, ask them what they need and want, then involve them in the design and build.

So that’s it. The five definitive points about AI for insurance technology. This may well be The Last Article You Need To Read About AI for Insurance. However, check back tomorrow. AI is advancing so fast, and hauling the entire global computing ecosystem along with it, that no article can be definitive for more than a short time.

All of the points Ralitsa Nenkova made in her excellent article remain valid as I reach the end of this diary entry. The tech has probably moved on since the beginning of the article, though, and looking back Point 3 is now smells a little stale. Readers, instead of keeping your legacy systems (as Nenkova and I have basically endorsed), now is probably the time when Agentic AI is sufficiently robust to allow a pain-free migration to a dynamic system that frees you finally from the shackles of legacy.

It’s Strategy 1, then. More than ever, you can and should dream about the best state your business could be in, the most amazing tools to which your employees could have access, and seamless, safe, real-time integration with all your trading partners’ platforms. It might not possible today, but likely it will be tomorrow.