We’ve moved past the “what if” of AI in insurance. Building on my previous article exploring the industry’s radical transformation, I’ll examine our current reality: where we are on this journey now, why we are here, and what needs to come next.
Make no mistake, insurers aren’t new to this game. They’ve been leveraging AI technologies for years. Fraud detection has capitalized on machine learning based systems. Underwriters have been automating decisions and bringing more flexibility to how premiums are priced. And various advanced analytics capabilities, provided by industry leaders like Aon, offer sophisticated AI tools that turn raw data into instruments for risk analysis.
In theory, no industry has more to gain from AI than insurance. In practice, no other sector has to deal with more complexity, risk, regulation, legality, and general compliance either.
Have we seen the use of AI in insurance change?
Yes, we have, and we are now seeing insurers move beyond “analytical” capabilities and chatbots into more advanced AI use in operations and services.
However, the reality of the widespread use of AI in this way is that when applied to insurance operations, we tend to see it being made to fit within the business model and legacy technology constraints rather than challenge them, creating AI-use fragmentation in point solutions. And perhaps worse – deepening legacy costs and dependency issues.
And now AI has got smarter, more capable of outputs, and it is moving into its agentic era where it can perform much more advanced tasks. Theoretically end-to-end across an entire lifecycle and across multiple different scenarios, processes, customer needs and experiences. This could exacerbate further.
Given this, the question I’m often asked at the moment is – are there any examples of agentic scale in insurance built on better foundations? And the short answer is not really; there are still only lots of pilots, PoCs and point solutions, which are either based on GenAI or using a form of agentic solutioning (so progress, but not at real scale).
The only exception to this tends to be in functions like developers, who are working in tandem with AI in ways that are arguably becoming a little more comprehensive.
And worryingly, perhaps the business use of AI can be advanced. However, this raises other AI concerns such as – does the CIO have the telemetry to manage, monitor and control all AI in their enterprise today?
While we aren’t yet seeing widespread adoption of products like the ServiceNow AI Control Tower, we are seeing a great deal of interest in it. Along with a shift to the use of AI to support GRC and audit processes – an area I think we will see a lot more appetite for during 2026-2027.
Within the context of what has therefore been modest AI driven change, and far too much pilot purgatory, insurance perhaps needs to refocus back on to how it can turn up differently in our lives forming the basis for the next batch of AI use cases.
At a recent roundtable I moderated with a room packed with senior insurance professionals, one example given was that of all the requests for AI (all meeting business case thresholds) only 4-5% had secured approval to go into production over the last six months. The general view was that we are running out of relatively low risk:high value ideas, and that now the risk tolerances and data models of insurers are starting to become a focus in the boardrooms. We are moving from demands to “just get AI out there” to calls now to “make sure it has an ROI and that it is implemented responsibly first”.
And perhaps this is a sign that rather than killing off AI in a wash of counteraction to the hype, we are instead moving into an era of making AI work responsibly, and that we now realize that this may be a little harder than some originally thought.
What is coming next on this journey?
I am proposing principles rather than outcomes to illustrate what is coming, because I think that the why and how of AI futures is more important at this stage than the what.
- Scale human-centric AI: I’ve always argued that the goal of technology should not be to replace human effort but to enhance it. I have written about the “sacred” nature of human struggle and the risk of losing our humanity if we outsource too much of certain thinking and thought processes to machines. I would equally argue that with AI there is the potential to extend human value, but we need models that constantly seek to ensure this is the case. Hope is not a strategy.
- Preserve the human voice: I often emphasize the importance of maintaining an “irreplaceably you” voice in a world where machines can generate competent but hollow sentences. In insurance there are vital moments when human to human interactions intensify in importance, AI can create more of these moments and make them more empathetic as well. The value of this is profoundly differentiating.
- Address the “intelligence gap”: Highlighting the gap between current AI capabilities and real-world context means noting that technology often lacks true context in order to be truly effective. We are now seeing agentic models that allow for more of this context in theory, but there’s a business model and data model problem that needs to be tackled in tandem.
- Evolve the enterprise design: When asked to put my “futurist” hat on, I am often also asked to discuss how AI is shifting industry operations from simple automation to intelligent agents and more sophisticated ecosystem thinking. I always need to make the point that this is a business model to shift more than anything. You need cross-functional teams in permanent collaboration and perpetual change. This is about creating an insurance enterprise that is constantly evolving.
- Tackle the data dilemma: Big data has the ability to change the way we see people, which impacts things like insurance and risk pricing. It should also change how we operationalize and use data across insurance as well. But there’s a problem. Legacy technology in insurance is rife, and most of the newer technologies in play mimic the way insurers work in those legacy systems. AI needs “good” data, but it also needs a new data model, well-orchestrated and made to work as near to real-time as needed. With all the deterministic parameters insurers will undoubtedly need.
- Make sure risk, security, regulation, compliance and legality are built in: If things are to move fast, these things will need to as well. So, working out how you build these into systems, business processes, and data models is essential. And I think that this all too often becomes an unlock to a lot more enterprise value than people realize.
The increasingly intelligent insurance future remains theoretically bright, so now we need to turn theory into reality and tackle the foundational issues in tandem with continued AI innovation. And this doesn’t make the potential for AI to create value any less likely; in fact it will make it even more feasible and viable. And in turn, it will likely extend our vision for AI across the industry.
