Insight banner

AI in Insurance Product Management: Insights from London's Leaders

Sofia Chaves
Sofia Chaves
Published: 13th November 2026
Last updated: 13th November 2026
AI in Insurance Product Management: Insights from London's Leaders

In November 2026, SGI hosted the Insurance Product Leaders Breakfast in London. Product leaders from across specialist underwriting, insurance broking, global insurance carriers and technology-enabled insurance solutions came together to discuss how AI is changing Product Management across the London insurance market.

Their view was clear: AI should speed up the work, but humans must stay in control of decisions, relationships and accountability. Here is what the room said about human oversight, getting real value from AI, build vs buy vs connect, and where the market should standardise.

Key takeaways

  • Use AI to extract, present and recommend. Keep humans in charge of decisions, client relationships and product roadmaps.
  • Make accountability explicit. That includes technology partners, before a major market event puts it to the test.
  • Judge AI against real human accuracy, not perfection. Expect adoption to take time.
  • Protect critical thinking and discovery skills, especially for people new to the industry.
  • Build where it differentiates, buy for speed, and connect where the market isn't ready yet.
  • Standardise the plumbing so firms can compete on product, pricing and risk appetite.
  • Bring brokers and underwriters in early and keep them engaged all the way to delivery.

 

Where should AI stop and humans take over in insurance?

AI should gather and present information. Humans should own the decisions that affect relationships, priorities or risk appetite. The group tested this against five everyday scenarios:

 

Scenario

AI or human?

Why the room chose this

Extracting data and triaging submissions

AI, with a human checking accuracy

Low risk. AI pulls data at pace and presents it, but no decision is made. A human confirms it is accurate, and in low-volume business a human in the loop adds even more value.

Declining a submission

Human control, with AI working within human-set rules

Declining is an active decision with relationship consequences. In a relationship-driven market, nobody wants an algorithm telling a broker "no". Some firms digitise their appetite and adjust the offer instead.

Writing the first draft of a PRD

Split views

Some felt the product manager must own the vision, or AI produces "slop" that is hard to unpick. Others found AI lifts drafts that were too thin for engineers to use, and challenges their thinking.

Deciding what goes on the roadmap

Human

Priorities change constantly, and some intangibles can't be reduced to a formula. AI doesn't yet have the forecasting, critical thinking or holistic view of product, customer and market that prioritisation needs.

Renewing a long-standing client

AI recommends, human owns the relationship

With claims history and a client profile, renewals are fairly deterministic, so AI can make a recommendation. The relationship between underwriter and client should stay with a person.

 

Submission triage was seen as low risk, because AI presents data and no decision is made. Declining business is different. In a relationship-driven market, nobody wants an algorithm telling a broker "no".

Leaders also warned against the “black box”. When AI declines business, a feedback loop should show underwriters what was turned away, so they don't lose sight of the market. Underwriters should keep control of the dials on appetite.

Roadmaps stay firmly human. Good product leaders need to understand the product, the customer and the market at the same time. The room felt AI doesn't yet have that holistic view.

 

Who is accountable when AI gets it wrong?

Today, humans are accountable. But the room agreed accountability isn't yet clearly defined as AI takes on more work.

  • Accountability sits with whoever sets the constraints. If someone controls the rules AI operates within, they must own the outcome.
  • Technology partners are a grey area. If a client sets a control and the technology lets through cases that don't meet it, contracts don't currently say where responsibility sits. There is no market-wide framework yet for the checks and balances AI needs.
  • Beware the rubber stamp. AI increases throughput. Nobody can meaningfully review 500 documents a day, so they just click "approve". At that point, human oversight isn't real.

The market hasn't faced a major event, such as a large hurricane, since these tools were widely adopted. When it does, the question of accountability will be tested.

 

Is AI solving real problems in insurance?

The room's answer: measure AI against how people actually perform today, not against perfection.

Customers often push back when AI isn't 100% accurate, but today's manual processes aren't either. One leader said that, when they last checked, people pulling information from proposals were about 83% accurate. The bar should be "as good as or better than us".

Two things are still holding the gains back:

  • Adoption takes time. Underwriters given AI summaries still want to see all the detail underneath, because they remain accountable for accuracy. The efficiency only comes once trust is built.
  • Cost is a risk. Building many processes on AI that is cheap today could backfire if costs rise significantly. Infrastructure teams are already planning for this.

 

How is AI changing the skills Product Managers need?

AI can bring people new to insurance up to speed much faster. But leaders worried it could also erode the skills that make good product people.

Experienced Product Managers know what good looks like because they did the work without AI. Newcomers may accept AI output because it sounds impressive and is the right length. Some may never have run a full discovery process.

The group's advice: train people to be critical of AI output. Used well, an AI first draft can sharpen thinking, because disagreeing with it draws out your own view. Several leaders felt the vision and direction should still come from the product manager.

 

Build, buy or connect: how should insurers decide?

Build where it creates a strategic advantage, buy when speed matters more, and connect where the market isn't ready yet.

Build gives strategic flexibility and helps firms scale. One firm builds most of its technology stack in-house, but still asks "can we buy this and get to market quicker?" every time. Brokers in particular hold rich client data up front, such as claims history, loss records and schedules of values. Buying a product ties that data to someone else's schema.

Building is hard, though. Creating a technology function that isn't corporate IT takes years, mistakes and long-term commitment. Systems built five or six years ago can become legacy that needs unpicking.

Buy is the lowest-friction choice if you have the budget. Firms with newer product teams and traditional IT often buy to deliver value quickly while they build capability. AI is changing the maths: some teams now run quick, low-risk proofs of concept to test how fast they could build before deciding to buy.

Connect makes sense while the market evolves. Placement platforms succeeded because they cleaned up unstructured data and gave brokers and underwriters a shared space. The longer-term direction may be direct API connections between brokers and underwriters. Until then, connecting gets value to customers faster.

 

Engagement matters as much as the technology

Whatever the route, success depends on bringing the business with you. Product teams that sit alongside underwriting turn "is this possible?" into "what's the trade-off?". A familiar pattern is excitement at discovery, silence during renewal season, then "that's not what I meant" at go-live.

 

What should the insurance market standardise?

Standardise the plumbing, and differentiate on underwriting.

The plumbing covers the data model, accounting, finance, claims and FNOL. This data moves between brokers, carriers and partners, so it can largely be standardised. Bordereaux show the cost of not doing so: a standard format exists, but every carrier asks for something different, and fixing that eats into time for innovation.

Differentiation should sit where underwriters have control: the product, pricing and risk appetite. That is how the London market wins business. The worst outcome would be insurance becoming a commodity.

The benefits of standards are practical. With a standard data model, one leader said, a new product can be delivered in four weeks rather than eight. Shared standards and consistent API connectivity can also bring down the cost of acquisition across the market.

 

How AI changes the standardisation debate

Earlier industry efforts focused on capturing structured data so it could be shared. With AI-powered submission ingestion, inputs no longer need to be as rigidly structured. The underlying data model still needs to be standard, to cut the time spent on data migration.

 

Who should drive it?

One view was that carriers should agree how they want to receive data and push it through their MGAs and delegated authority, with brokers' tech teams following. As one leader put it, "let underwriters underwrite", then turn their output into the format carriers need.

Brokers on the ground may not care about standards. But many are now seeing the cost of placing business in London as a threat and are investing heavily in digital connectivity. That makes them more willing to engage.

 

Frequently asked questions

How are insurance product leaders using AI?

They mainly use it to extract data from submissions, triage incoming business, support first drafts of product documents, and make recommendations, such as on renewals. Decisions that affect clients, priorities or risk appetite stay with humans.

 

Should AI be allowed to decline insurance submissions?

The leaders were cautious. Declining is an active decision with relationship consequences. Some firms digitise their appetite and adjust offers within human-set rules instead of declining outright. They also give underwriters visibility of what was turned away.

 

Who is accountable for AI decisions in insurance?

Humans remain accountable today, particularly whoever sets the rules AI works within. Accountability between insurers and technology partners isn't yet clearly defined in contracts, and the market has no shared framework yet.

 

Should insurers build or buy technology?

Build where it creates a competitive advantage, buy when speed matters and capability is still developing, and connect to existing market platforms where the market isn't ready for a new model.

 

What should the London insurance market standardise?

The underlying "plumbing": data models, accounting, finance, claims and FNOL. Firms should differentiate on product, pricing and risk appetite.

 

Join the conversation

This was the first Insurance Product Leaders Breakfast, hosted by Senior Consultant Sofia Chaves, and we're keen to grow the network and bring more perspectives to the table. If you're an insurance product leader and would like to join the next conversation, get in touch with Sofia here.