Marrow

The AI Distribution Playbook for UK Insurers · 2026

Your customers are asking AI where to buy insurance. Are you in the answer?

Research, infrastructure, and a practical view on what UK insurers should do as AI assistants become a primary route to discovery.

m.

Marrow

Research and analysis by the Marrow research team. Methodology: 200+ queries across four AI platforms, March and April 2026.

There is no shortage of writing on AI in insurance. Most of it sits at the level of frameworks and forecasts. This document takes a different angle: what AI assistants are actually saying about insurance products today, where the gaps are, and what the practical response looks like for an insurer's CMO, CTO, and compliance team.

In March and April 2026, the Marrow research team ran more than 200 insurance-related queries across ChatGPT, Claude, Perplexity, and Gemini. The queries covered motor, home, travel, life, and pet, framed the way real customers ask them: "what's the best home insurance for a first-time buyer?", "recommend car insurance with breakdown cover", "I'm 24 and just bought a flat in Leeds, what should I get?". We then tracked which insurers appeared in the responses, how often, and with what accuracy. The picture varies by line and by platform. The direction of travel does not.

On this page

  1. 01Who this is for
  2. 02The next distribution shift
  3. 03How buyers find insurance now
  4. 04Stakeholder alignment
  5. 05Why direct AI integration is risky
  6. 06What the right infrastructure looks like
  7. 07Four phases of readiness
  8. 08Questions to ask right now
  9. 09Your 90-day action plan

01

Who this is for

This is for senior leaders in an insurance business who can see the platform shift happening, but don't yet have a clear picture of what it means for distribution.

CMO

You own customer acquisition and your brand is largely absent from AI answers

You've built the comparison site strategy, the SEO playbook, and the paid media engine. A new channel is emerging that your team has no way to participate in yet, and the visibility gap compounds the longer it stays open.

CTO

You've been asked to have an AI strategy and don't want to build it wrong

Direct LLM integration sounds appealing but introduces compliance and underwriting risk that legal teams typically won't sign off on. You need an architecture that works without changing core systems.

Head of Distribution

You're watching new channels emerge that you have no access to

Your job is finding new routes to customers. AI assistants are becoming a primary route, and most insurers have no presence in those environments today.

Outside those titles: if you are responsible for how your insurance business reaches customers in the next five years, this is for you.


02

The next distribution shift

Every generation of insurance distribution is defined by the platform customers use to find products. The 1990s belonged to the phone book and the broker's desk. The 2000s, to the search engine. In the 2010s, aggregators such as MoneySuperMarket, Compare the Market, and GoCompare became the dominant intermediary. Each shift created winners and reshaped the economics for everyone else.

The insurers who won comparison sites weren't the ones with the best products. They were the ones who got their data architecture right first.

AI assistants are becoming the interface through which a growing proportion of customers discover and evaluate insurance products. The channel already exists; the question is whether your products appear in it.

Hundreds of millions

Weekly users across ChatGPT, Claude, and Gemini combined, with a growing share asking insurance queries

<2%

of UK insurers can currently transact natively through an AI assistant

Across the queries we ran, most major UK insurers did not appear in the answers. Of those that did, the conversation typically ended at the brand mention, with no path through to a live quote.


03

How buyers find insurance now

A growing share of customers no longer start their insurance journey on a comparison site. They start with a conversation: "ChatGPT, what's the best home insurance for a first-time buyer?". The behaviour is most pronounced among digitally native buyers, and it is moving up the age curve faster than most teams expect.

The old funnel

Step 01

Awareness

TV, outdoor, and paid search build brand familiarity. The customer already knows your name before they shop.

Step 02

Consideration

They visit an aggregator. Your price and features compete on a grid they can see and sort.

Step 03

Purchase

They click through to your site or buy via the aggregator. Transaction complete.

The new funnel

Step 01

Conversation

The customer asks an AI assistant what they need. The AI recommends two or three options. If your brand isn't in those recommendations, the funnel never reaches you.

Step 02

Instant qualification

The AI agent asks qualifying questions, collects details, and structures the risk inside the same conversation.

Step 03

Transaction

The AI passes structured data to your infrastructure, retrieves a live quote, and the customer binds, without ever visiting your website.

What's different

In the old model, you competed on price once the customer arrived. In the new model, you compete on discoverability before they arrive. Visibility becomes a prerequisite for the sale, not a step in the funnel.


04

What your stakeholders need to hear

Most of the work in moving on this is internal alignment. Different stakeholders come at it with completely different questions.

StakeholderWhat they're worried aboutWhat they need to hear
CMOLosing brand equity in a channel they can't control or measureThis isn't brand dilution. It's the next aggregator. You either have a feed into it or you don't.
CTOIntegration complexity and legacy system exposureThe infrastructure sits in front of your existing APIs. No core system changes. One integration unlocks all AI channels.
ComplianceFCA risk, AI hallucination, auditability of AI-driven decisionsDirect LLM integration is the risk. Compliance-ready middleware is the solution. Every transaction is auditable end to end.
Head of DistributionChannel conflict with existing broker and aggregator relationshipsAI is an additive channel. It surfaces customers who would never have come through your existing routes.
CFOROI justification for a new infrastructure investmentGWP at risk if competitors move first is measurable. The question is whether you want to be first mover or catch-up.

The most important internal dynamic: this decision is typically championed by a CMO or Head of Digital, then evaluated by the CTO. Both need to be convinced. The CTO's concerns about legacy integration are legitimate and addressable; so are the compliance team's. Naming them directly is a faster route to alignment than working around them.


05

Why direct AI integration is risky

The obvious move, integrating directly with an LLM provider's API to build something in-house, carries risks that most insurance technology teams haven't fully mapped.

The hallucination problem

Large language models are probabilistic. They predict what a plausible answer looks like, rather than calculate what the correct answer is. In most domains a slightly wrong answer is tolerable. In insurance it isn't: if an AI assistant tells a customer they're covered for something they're not, that is a liability problem, and potentially an FCA one.

AI is probabilistic. Insurance is deterministic. That gap is where compliance risk lives.

The missing data structure problem

Insurance underwriting requires structured data; AI conversations produce unstructured data. The customer says "I live in a Victorian terrace in Leeds." The underwriting system needs "Property type: terraced. Construction: stone. Year built: pre-1900. Location: LS postcode." Translating reliably between those two, at scale and in a compliant way, is non-trivial.

The audit trail problem

FCA regulations require that insurance sales processes are auditable. A direct LLM integration typically produces no audit trail by default.


06

What the right infrastructure looks like

The solution is a compliance layer that sits between the AI assistant and your existing systems. It handles translation, compliance, data structuring, and the audit trail, so your core systems do not have to change and the AI platform does not have direct access to your underwriting engine.

It is the same idea as a payment gateway. Stripe doesn't rebuild your banking relationship; it sits in front of it and handles the complexity so the product team doesn't have to. The same pattern applies to AI distribution.

Layer 1

AI Platform

ChatGPT, Claude, Perplexity, Google AI. The customer conversation happens here. This layer asks qualifying questions, collects risk data, and constructs a structured query.

Layer 2

Marrow: Compliance-Ready Middleware

Validates and structures incoming data, generates the audit trail, and routes the structured request to your pricing engine. Returns a compliant, formatted quote. Designed to support FCA compliance requirements throughout.

Layer 3

Your Existing Systems

Your pricing APIs, policy management system, and underwriting rules. These don't change. One Marrow integration unlocks all AI distribution channels simultaneously.

A single integration covers every AI platform, with no core system changes, the same way one Stripe integration handles every payment method.

This is also the answer to the CTO's question about maintenance. A direct integration with ChatGPT requires a second one for Claude, a third when Gemini becomes dominant, and a fourth when the next challenger emerges. One middleware integration updates centrally as the AI landscape shifts.


07

The four phases of AI distribution readiness

There is no single go-live moment. AI distribution readiness builds in phases. Sequence matters more than speed, though speed matters more than most insurance businesses currently appreciate.

Phase 1

Visibility

Your brand appears in AI-generated answers to insurance queries. You're in the conversation, even if you can't yet transact.

Phase 2

Discoverability

Your product information is structured so AI systems accurately represent your products, pricing range, and eligibility.

Phase 3

Transactability

A customer can receive a live, compliant quote through an AI conversation and bind a policy. The full transaction happens in the AI environment.

Phase 4

Optimisation

You are measuring AI distribution volume, optimising product data for AI performance, and expanding across platforms as they grow.

Most major UK insurers sit somewhere between Phase 0 and Phase 1 today. A small number, including Aviva, are actively investing in AI distribution capability. The gap between Phase 1 and Phase 3 is not a technology gap; it is an infrastructure decision.

Phase 4 is where compounding begins. Insurers operating here can see which product descriptions convert best, which qualification flows reduce drop-off, and which AI platforms produce the strongest risk profiles. That data advantage is hard for late movers to buy back.


08

The questions you should be asking right now

The visibility question

Ask someone in your team to open ChatGPT, Claude, and Perplexity and run ten insurance queries relevant to your product lines. Count how often your brand appears. Count how often your top three competitors appear. The gap is the business problem.

The integration question

Ask your CTO: if we wanted to make our pricing API available to an authorised AI distribution partner, what would that require? If the answer is "six months and a core system change," the constraint is the architecture assumption, not the technical reality. A well-designed middleware layer can connect to a standard pricing API in weeks.

The compliance question

Ask your Head of Compliance: what would an AI-assisted insurance sale need to look like to satisfy our FCA obligations? This is a solvable problem. The answer is an audit trail, clear disclosure of the AI's role, and a compliant data collection layer, all of which can be built into well-designed AI distribution infrastructure.

The first-mover question

Ask your leadership team: if our largest direct competitor launches AI distribution in Q3, what does that mean for our new business pipeline? Insurers who were fastest to get their pricing onto aggregators in 2004 to 2005 captured disproportionate volume for the next decade. The ones who waited until 2008 spent years buying back market share they had given away.

The honest answer

You don't need to be first. The historical pattern is clear: the window between first-mover advantage and category normalisation in distribution channels is typically 18 to 24 months. Insurers who acted on aggregators in 2004 to 2005 captured disproportionate volume for the next decade.


09

Your 90-day action plan

Here is what we'd recommend for any insurer who has decided this is a priority.

Days 1 to 14

Audit your current AI visibility

Run a structured audit of 50 to 100 insurance queries across ChatGPT, Claude, Perplexity, and Google AI. Document which brands appear, at what frequency, and with what accuracy. Benchmark against your top five competitors. This is your baseline and your internal business case.

Days 15 to 30

Map your existing API infrastructure

Identify your current pricing API endpoints: what data they accept, what they return, and how they are authenticated. These don't need to change. The aim is to understand what a middleware layer would connect to. A CTO-led conversation that typically takes a week.

Days 31 to 60

Run a pilot on one product line

Don't try to solve all of insurance distribution in quarter one. Pick one product line, typically motor or home, and build the pilot on that. Define the qualifying questions, map them to your underwriting data requirements, connect the infrastructure, and test end to end in a sandbox environment.

Days 61 to 90

Go live and measure

Move the pilot to live with a defined scope. Instrument your measurement from day one: query volume, quote completions, bind rate, policy data quality. The data from the first 30 days of live operation will tell you more than any pre-launch planning session.

Ninety days from kickoff to live is achievable with the right infrastructure in place. The bottlenecks are almost always internal alignment and compliance sign-off, not technical complexity. The teams that move fastest treat this as a priority with a senior sponsor, rather than a project.

© 2026 Marrow. Marrow is a trading name of Project Malcolm Limited.