Novee’s event on 4 March was standing room only. But the turnout wasn’t the most interesting part. The conversation was.
The most credible voices in the room were not talking about AI in insurance as a future ambition, a transformation programme, or a flashy demo. They were talking about tools already changing underwriting work today.
That shift in tone is worth paying attention to.
Insurance has spent years discussing digitisation in broad terms. The language is often expansive, but the operational reality is narrower and harder. Underwriters still work through dense submissions, fragmented documents, legacy workflows and acute time pressure. In specialist lines, that challenge is even more pronounced. Valuable risk information is often buried deep inside surveys, schedules of values and attachments.
Teams do not lack data. They lack time, structure and context.
AI in insurance, grounded
What came through clearly in the panel discussion is where value is starting to emerge. Not from replacing underwriters, but from helping them process more of the information in front of them, surface what matters faster, and make better decisions with more confidence.
Alex Priestley, Head of Power & Renewables at Lancashire Insurance, described the reality plainly. Complex engineering submissions can contain 50 to 100 survey documents, often running to 75 or 100 pages each. In practice, teams may only be digesting around 10% of that material.
“It’s death by a thousand tasks, we needed to reduce that and get back to focusing on deal-making and negotiation.”
That framing is useful. It shifts the focus onto removing friction from the underwriting process. The objective is not to replace judgement, but to support it under real-world constraints.
Targeted, not sprawling
Kevin Cleary, COO at Volt Underwriting, captured another familiar challenge:
“Two words that scare the hell out of me: digital transformation.”
His point was less about the concept itself and more about how it often plays out. Large programmes can become so broad that they lose operational focus. Everything is in scope, and progress slows.
The alternative he described was more targeted: break problems into smaller pieces, focus on a specific bottleneck, and solve for efficiency and accuracy in a way that creates tangible value.
That approach seems to be resonating.
Alex explained that Lancashire had considered a number of AI solutions, many of which required significant investment, long implementation timelines and deep integration into existing systems. What the business needed was
“more modular, more plug-and-play, more bespoke. Something that could deliver the information underwriters need to make better decisions, faster.”
With Novee, the process from initial engagement to proof of concept took around six weeks, followed by two weeks of testing. “We got the value we were looking for,” he said, which made it possible to take a business case to the C-suite for approval.
In the current environment, that speed of validation is notable.
Context over extraction
There is also a broader product shift taking shape.
Much of the early wave of AI in insurance has focused on extracting data from unstructured documents. That remains useful, but it is no longer sufficient on its own.
As Haris Khan, CEO and Co-Founder of Novee, put it:
“Extraction is table stakes. Context is going to be everything.”
The distinction matters. Pulling data out of a submission improves efficiency. Understanding what that data means — in the context of a portfolio, a market, or a specific underwriting strategy — is where more meaningful value begins to emerge.
Underwriters are not just asking what is in a submission. They are asking what it says about risk selection, how it compares to existing exposures, and whether it should be written at all.
That shift from extraction to interpretation came through consistently in the discussion. Carol Baker, Head of Digital Strategy at Liberty Specialty Markets, described how the focus had moved beyond process efficiency towards risk selection. Helping underwriters identify which risks to prioritise, rather than simply processing more of them.
Better decisions, faster
The most convincing signals still come from how these tools are used in practice.
Alex pointed to improvements in data quality, including better geocoding and additional information that would previously have been missed because it was buried deep in documentation. He also noted that
“our loss ratios have dropped at both a risk and portfolio level.”
That kind of outcome is what ultimately matters. Speed is important, but only if it is paired with better decisions.
Carol described a similar pattern at Liberty, where underwriters across regions had begun sending submissions through and “raving about the value they’re getting back… particularly because of the speed of turnaround.” At the same time, she emphasised that the goal is not to remove underwriters from the process.
“We want to give them as much insight as possible so they can make better decisions.”
That balance feels important. In specialty underwriting, expertise is not easily replaced. But it can be better supported.
What this suggests
There is a growing narrative that value will come from large-scale transformation and a wholesale reimagining of underwriting. That may happen over time, but it is not how progress is unfolding today. What we are seeing instead is more incremental: focused applications that fit into existing workflows, solve specific problems, and deliver value quickly.
What feels different today is the demand. Underwriters are actively calling for better digital and AI-enabled ways of working, driven by rising submission volumes, more complex data, and pressure to respond faster.
That shift creates a clear window of opportunity.
The risk is slowing progress by tying change to multi-year transformation programmes. The opportunity is to deliver value within weeks, in the flow of underwriting, and build from there. This is where decision support becomes central, helping underwriters interpret information, prioritise effectively, and act with greater confidence.
There is still a long way to go. As Haris noted, building something that works in production, day in and day out, requires a different level of rigour. But the direction is becoming clearer. The most valuable applications of AI are emerging through these practical, incremental wins.
Progress is more likely to come through evolution than revolution.