Insurance is a chore. Chores get automated first.

The FCA's Mills Review puts insurance last among financial products for consumer AI use today. By 2030, we think it will be first.

The Review's prediction is that retail finance becomes AI-enabled, continuous and delegated. Its chart of progress so far sets how many people shopped for a product last year against how many used AI to help. Debt advice sits top left: sought by 2% of adults, nearly half of whom turned to AI. Insurance sits in the opposite corner: one in three of us bought or switched in the last twelve months, but only 8% used AI to help.

This makes sense. The products at the top of the chart are the less familiar ones. Debt, pensions, investments, mortgages: decisions people face rarely, with consequences that may run for decades, and often little instinct for whether they are getting them right. People want to understand those before they act, and AI is ready to help.

However, insurance is more routine: it is fairly commoditised, and bought through a funnel the industry has spent twenty years optimising. When the comparison is on price alone, and for many people it is, there is little for AI to add.

Managing your investments feels like a thesis-driven decision that you want to be across; managing your insurance feels more like a chore that you would rather put off.

Assistance is lowest exactly where delegation will be highest

We believe insurance is about to climb this chart faster than anything else on it. Not because people take a sudden interest in their insurance, but because AI is about to be able to handle the chore for them almost end to end.

The Review maps adoption along five levels of autonomy. The chart above measures where we are today: level 1 and 2 behaviour where the human leads, and AI supports or collaborates. From level 3 the roles reverse. AI leads, and the human guides it, then approves its actions, then simply monitors the outcomes.

Products people want to deeply understand and be opinionated about will climb those levels slowly. Others will climb through the levels quickly.

In other words, the low level of AI assistance seen today might actually be a signal that the AI adoption curve will be the steepest as we head towards delegation.

When that happens, insurance will still be an outlier, except at the other end of the adoption spectrum.

From optimising customer experience (CX) to optimising agent experience (AX)

When that climb comes, distribution moves with it. AI, the Review says, "may not remove intermediation so much as relocate it", towards agents, platforms and the data layers beneath them.

That shifts the discipline. Twenty years of work went into customer experience: fewer questions, faster quotes. What matters next is agent experience, whether a machine acting for your customer can find your product, read it and act on it.

Products, the Review notes, may need to be presented "in ways that are legible to those systems as well as to humans". Anything an agent can't read reliably, it treats as absent. In an agent-led market, an unreadable product is an unavailable one.

That is why we are building the Global Product Graph: policy documents and market information in, structured machine-readable product data out. This is the foundation that will allow consumers to put their insurance on auto-pilot, with smart switching and renewals handled inside limits the customer sets.

And as agents become your customers, being able to show what happened matters more, not less. That is what the AI Safety Harness is for: guardrails on every conversation as it happens, and an audit log for afterwards. Today it powers AI Companion, live inside SuperSaveClub Insurance, the digital broker we launched with MoneySuperMarket. Tomorrow it will power the agent-led broking that replaces the customer-facing funnel.

If you are working out where your customers will start their journey in 2030, we would love to talk.