Aviva's Building a ChatGPT App. Claims Teams Are Holding the Line. Both Are Right.

For the sector application, see our AI for insurance broking guide and 5 Pillars of AI for insurance brokers.

Aviva announced this month it is launching the UK's first ChatGPT insurance app. Simply Business added ChatGPT to its SME quoting flow. CFC kicked off an agentic AI pilot for underwriting submissions. The same month, the industry press ran pieces titled "Why AI should never make claims decisions" and "Tech must enable empathy in claims".

Both of those positions are right. Understanding why is the most useful thing service SME operations can do this quarter.

What's actually happening

The pattern across the last few weeks of insurance press is consistent. Distribution is moving fast. Claims is moving carefully.

On distribution, Aviva is going first on a consumer ChatGPT app. Simply Business is using ChatGPT to lift SME quote conversion. CFC's "Lane Assist" pilot is turning broker emails into quote recommendations. One insurer told Insurance Post it has seen a fourfold conversion uplift from a ChatGPT-driven channel.

On claims, the LMA has published an AI adoption toolkit. The FCA confirmed it will regulate AI through existing fairness and consumer duty frameworks rather than building a bespoke regime. Zurich's claims lead reminded the industry that technology has to enable empathy, not replace it. A separate Post piece flagged that consumers still want humans, not AI, when claims get complex.

Same month. Two very different speeds.

Why the gap exists

It isn't because the technology is harder for claims. It's because the work is shaped differently, and the cost of getting it wrong is asymmetric.

Quoting is textbook automation territory. Most of the rules are explicit, the decisions are bounded, and a bad quote loses you a sale, which is recoverable. A bad claims decision can cost a customer their car, their settlement, or their trust in the entire industry, and it can earn you a regulator visit you didn't want.

That asymmetry is the whole story. Claims teams aren't behind the curve. They're operating in a different risk environment, and the FCA's stance reinforces it.

Where the 3Rs framework actually applies

The 3Rs filter (Repetitive, Rules-based, Resource-intensive) is how we cut through the noise in every Optimus discovery meeting. Run any task in a service operation through it.

FNOL triage, intake and document classification. Three out of three. Build here first.

Claims correspondence, TPI rebuttals, BHR challenges, hire rate arguments, settlement nudges. Three out of three. Patterned, evidence-driven, high volume. This is where credit hire and motor claims automation earns its keep right now.

Fraud signal detection and return-to-work pattern analysis. Three out of three. Pattern recognition over a back-book is exactly what these models are good at.

Final liability decisions, complex injury assessments, vulnerable customer handling. Not all three. Resource-intensive, yes. Fully rules-based, no. So you augment, not replace. The handler stays in charge. The AI does the admin around them.

This is the asymmetry the news is showing in slow motion. Distribution is doing the full three-tick work. Claims has a long list of three-tick tasks too, but they sit in the admin layer, not the decisioning layer. SOS Operations is where the value is hiding.

What this looks like on a real desk

A handler picks up a credit hire file in the morning. The TPI has written in challenging basic hire rates. There's thirty minutes of work ahead. Pull comparator rates. Cross-check the claimant's profile. Find the relevant case law. Draft a response. Diarise the chase.

In a 3Rs-shaped operation, the model does the heavy lifting and the handler reviews. The comparator pack is built. The case law is sat in the file. The draft letter is in the inbox already, ready to edit. Thirty minutes compresses into five. Handler decisions all the way, but with the evidence and the admin done.

That's the operational pattern hiding behind every distribution headline. Same logic, applied a few layers further back in the value chain.

A fair counterpoint

The distribution sprint isn't risk-free either. A ChatGPT quoting flow that misrepresents coverage is a mis-selling problem with the FCA's name on it. A consumer app that confidently hallucinates a policy benefit is going to cost someone their renewal. The same asymmetric-risk logic that's keeping claims cautious will eventually catch up with distribution too. Anyone building at the front of the funnel right now should be planning for that, not assuming the regulatory weather will stay favourable.

What to do this month

1. Run your top five handler tasks through the 3Rs filter. The three-out-of-three ones are your first build, not your last.
2. Separate decisioning from admin. Automate the admin and evidence-gathering. Leave decisioning with the human and document the audit trail.
3. Read the LMA's AI adoption toolkit even if you're not in the Lloyd's market. It's the most practical UK governance template published this year.
4. Get your claims data clean. Without clean data, AI returns shrink fast and audit conversations get awkward.

Where to start

If your AI roadmap currently lives on a slide, pick one process this month. Probably claims correspondence. Probably FNOL triage. Build the smallest possible version with a handler in the loop. Measure the time saved per case. Show your team the result. Then go again.

The distribution layer of insurance is already running. The claims layer is walking, for good reasons. The teams that win the next twelve months are the ones who pick the right tasks to automate in claims, not the ones who try to automate the whole desk.