There's a number that should embarrass the travel insurance industry: roughly 80%.
That's the average claims acceptance rate. Which sounds reasonable until you flip it. One in five claims is rejected. And in most of those cases, it isn't fraud. It's a customer who thought they were covered, because nobody clearly told them they weren't.
The FCA noticed. In December 2025, responding to the Which? super-complaint on travel insurance, the regulator pointed to systemic issues around sales practices and what it called the "Consumer Understanding Gap." The chasm between what customers think they're buying and what they've actually bought.
This is the problem we've spent the past few months prototyping a solution to. It's called CarriffAI.
What's Actually Going Wrong
The failures are frustratingly mundane. A single-item baggage limit of £300, but the customer's camera is worth £900. A gadget add-on that excludes devices over 5 years old, but the customer's iPhone is 6 years old. A skiing trip where off-piste activity isn't included, but no one thought to ask.
None of this is hidden. It's all in the policy wording. The problem is that policy wording is written by lawyers for regulators, not by humans for humans. When a customer has 4 minutes and a comparison site, they're not reading pages 10 to 12 of a PDF.
Traditional brokers address this inconsistently. Call centres vary by advisor, by shift, by how many calls are queued. And there's often no auditable record of what was actually said.
The CarriffAI Approach: Digital Steward, Not Digital Salesperson
We've built a working prototype of what we call a "Digital Steward." An AI agent whose job is not to sell, but to make sure the customer genuinely understands what they're buying before they buy it.
In practice, that means three things.
First, proactive gap detection. Before surfacing any policy, the AI checks for the most common coverage failures: single-item limits, gadget age conditions, trip duration excesses, potential overlap with home insurance or bank account cover. It flags these issues unprompted, not because the customer knows to ask, but because the system knows to check.
Second, itinerary parsing. Upload your TUI booking confirmation or holiday PDF and the AI reads it, extracting the destination, travel dates, traveller names, and trip cost, then pre-fills your quote automatically. No typos, no wrong dates, no drop-down errors.
Third, evidence-first claims thinking. We want to start collecting proof of purchase and key policy details at the point of sale, not at the point of claim. This directly targets one of the FCA's specific concerns about the repetitive information requests that delay or derail legitimate claims.
Where We Are
We want to be honest about what CarriffAI is right now: a prototype. A working, demonstrable concept, but not yet an authorised firm.
We've applied to take part in the FCA's AI Live Testing Cohort, the regulator's own sandbox programme for responsible AI innovation in financial services. We believe that's the right route, building the regulatory relationship properly from the start rather than retrofitting compliance later.
If accepted, the cohort gives us the framework to test our model with real users under FCA supervision, and to demonstrate that AI-driven insurance advice can produce better consumer outcomes than the current standard.
We're not sharing a public link to the prototype. We'd rather walk you through it personally, because the context matters. The demo is most useful when we can explain what you're seeing and why each design decision was made with Consumer Duty in mind. If you'd like to see it, get in touch and we'll set up a call.
The Technical Foundation
The prototype runs on GCP, with the agentic reasoning layer built on Vertex AI and document parsing handled by Document AI. The interface is conversational, a chat flow that feels like talking to someone who actually knows the product rather than filling in a web form. Behind the conversation is structured logic, branching risk assessment, and a full interaction log designed from the start to satisfy FCA outcomes monitoring requirements. Data architecture is UK-based throughout, aligned to UK GDPR with no US infrastructure.
What We're Looking For
An insurer or MGA who wants to be the underwriting partner behind a Consumer Duty-native distribution platform, built specifically around the gap the FCA identified. Or an investor who sees what we see: a £2bn+ market, a regulator actively demanding better outcomes, and no clear AI-native winner yet.
We're not building a chatbot layer on top of an aggregator. We're rethinking the sales and advisory process from first principles, with the FCA's own concerns as our product brief and a regulatory testing pathway already in motion.
If that sounds interesting, we'd love to show you what we've built.
Get in touch at engage@carriffgroup.co.uk
CarriffAI is a pre-authorisation prototype. It does not currently hold FCA authorisation and does not sell or arrange insurance products.