When InsTech and AI Risk first ran this event in November 2025, the audience came to work out what Agentic AI was. This time the framing had shifted. Boards want investment cases that survive scrutiny, technology leaders are being asked to move past disconnected pilots, and underwriting, claims and operations teams want to know how the work itself changes. The day was built around that harder conversation.
Where the value sits
Simon Torrance, founder and CEO of AI Risk, set the spine for the day in his opening keynote. The tools on show, he argued, give only a temporary advantage: competitors can buy the same platforms, models and vendors. What compounds instead is what he called “intelligence capital”, the accumulated reasoning behind how a business underwrites, handles claims and qualifies opportunities, captured so it can be reused and scaled.
Torrance’s point was that most AI spend today buys productivity and reaches parity, while judgement remains the real battleground. He pointed to a business he helped build two years ago that runs largely on a team of collaborating agents; it has been operating with a loss ratio around half the industry average for its product. His practical challenge to the room: own the coordination layer and the memory of how decisions are made, rent the models and start now. Fast-following, he argued, is the weakest strategy in agentic AI.
That thesis recurred all day. Ashok Krishnan of AXA XL described building an “information capital” layer on top of vendor tools. Erdal Atakan of Inigo talked about codifying a company’s tacit knowledge as its moat. Manuela Veloso returned to the same idea from the banking side in her session later in the day.
From pilots to production
Three insurers showed what commitment looks like in practice. Alice Jin, Agentic and generative AI lead at AXA UK, walked through the company’s first agentic solution, live for a few weeks: eight agents handling higher-value third-party motor injury claims end to end, from document ingestion to supporting arguments, with a human review at the end. It delivered around 85% efficiency saving on the process and took nine months from idea to production. Her advice was to bring end users in from day one and treat them as collaborators, not obstacles.
Pieter Viljoen, chief data officer at Allianz Partners, described two years automating claims across 70 countries on a platform of roughly 270 reusable components. His counsel was blunt: don’t think in use cases, think platform and capabilities, tackle the hardest problem first and put a multidisciplinary product team around it rather than a project. Some claims, he noted, are now paid in around 60 seconds.
Tjeerd Bosklopper of NN Group, further along than most, made the case for a strategy that connects cost, growth and customer experience rather than a queue of separate initiatives. His firm chose an industrialisation route, fixing legacy foundations first, and over-invested in re-skilling its top leadership, which he said mattered more than expected. On the London Market panel, Atakan and Nick Williams-Walker of McGill & Partners added the operational reality: adoption depends on trust, sometimes on showing people exactly how an agent is built and increasingly on the token costs now appearing in the P&L.
Economics, governance and the frontier
The economics question ran throughout. Amélie Breitburd presented EuroDIEM, an initiative to industrialise syndication and open up an estimated €200 billion of new European premium by making hard-to-place risks such as cyber, climate and pandemic more insurable, expanding the market rather than optimising the existing one. Elena Maran of Alethesis AI argued that governance has to move from the system level to the individual decision, introducing “runtime authority” that scores each agent action for risk and routes it to a competent human only when warranted.
The frontier sessions were designed to stretch the room. Arnaud Vincent of Swiss 6022 demonstrated self-organising agent “swarms” and digital twins, including one that autonomously ordered him a pizza with money it had earned. Manuela Veloso, who founded and led JPMorgan Chase’s AI research for eight years, reminded the audience that finance and insurance come down to numbers, documents and people, that agents should log and reuse what they learn, and that the honest advice is simply to start with simple things done well.
Recognising the practitioners
Between sessions, the inaugural Top Agentic Faces in Europe awards, hosted by Ian Thompson and supported by Otera, recognised five people moving Agentic AI into production. Winners included Pieter Viljoen of Allianz Partners for transformation at scale, Anthony Elliot of Zurich Insurance for an Agentic AI hackathon that drew 1,000 participants across 40 countries and Ed Klinger of Flock for rethinking insurance through telematics-led fleet cover, recently recognised by an £80 million acquisition by the Admiral Group. Klinger’s takeaway captured the mood: the best claims experience is no claims and the industry should use AI to help customers avoid losses, not just process them faster.
The view to 2030
The closing panel resisted talk of a “job Armageddon”. Bosklopper expected a continued, gradual decline in headcount rather than a cliff, with humans and agents together doing the work. Ed Ackerman of Qover made the case for embedded insurance, which he sized at around $70 billion USD last year and potentially ten times that by the end of the decade. Breitburd and the panel returned to the protection gap, and to the possibility that better economics ultimately flow back to the customer. As Bosklopper put it, the prize is more coverage, better service and more people insured at today’s margins, leaving the customer as the winner.
The next 12 months
Robin Merttens and Simon Torrance closed by pulling the threads together. If November 2025 was about understanding what Agentic AI is, this event was about deciding what to do with it. The recurring message across the day was a consistent one: the advantage will not come from the tools everyone can buy, but from the reasoning a business captures, the foundations it fixes and the discipline it brings to deciding where AI is worth using at all. On the evidence in the room, the firms already treating that as a strategy rather than a set of experiments are the ones with something to show for it. The next 12 months will tell how far the rest of the market follows.
Join us for next year’s event on 14 September 2027 at 1 Basinghall Avenue, London EC2V 5DD. Save the date now and we look forward to welcoming you.
