The choices that will define your AI customer journey

Building on our previous insight, The customer journey is being remade by AI, where we explored how AI is reshaping the insurance customer journey and the outcomes now taking shape across the market, the focus now turns to what drives those outcomes in practice. They are not the result of technology alone, but of a series of decisions that shape how AI is designed, deployed and governed. Some are deliberate, others emerge over time, but together they determine the experience delivered to customers and the risks firms carry. Understanding these choices is key to moving from passive adoption to intentional strategy.

Three choices that determine which state your firm reaches

How does a firm end up at Default when nobody chose Default? Three choices, made at every moment of the customer journey, determine where a firm lands.

Choice one: Who AI serves

Combines two related questions. At the pricing level: how granular do we let the model go before precision becomes exclusion? At the investment level: where does AI capital go, optimising the existing book or extending the market? Both answer who gets to be a customer.

Precision means better selection, lower loss ratios. Existing-book optimisation has measurable ROI; market extension does not. The gradient pulls toward sharper pricing of existing customers and away from extending to new ones. For example, while Marshmallow used alternative-data AI to insure one million previously excluded drivers profitably at a USD 2 billion valuation, no incumbent has copied the model.

“At what point does pricing precision stop being actuarially justified and start being social sorting, and can you point to that boundary in your model?”

Choice two: What AI hides

What can the customer see about how decisions are made? Where does the firm's information advantage become an information asymmetry the customer cannot navigate? The right to challenge under DUAA 2025 is meaningless if the customer does not know what to challenge.

Information advantage is a competitive moat. Visibility costs money and creates audit risk. This choice pulls toward more sophisticated AI and less customer-facing visibility into its workings.

For example, telematics pricing has been documented making consequential premium decisions on inaccurate data the customer could not see. Renewal pricing is optimised on behavioural data the customer cannot know exists.

“Does your AI improve outcomes for consumers or improve extraction from them, and do you have a framework that can tell the difference at model level?”

Choice three: Who answers for AI

When the model is wrong about an individual, who decides? When SM&CR is invoked, who is named? Where do humans sit in the decision loop, and are they real reviewers or nominal signoffs?

Removing humans produces faster decisions and lower cost-to-serve. Keeping them in catches edge cases at cost. The gradient pulls toward less human review. The DUAA 2025 gives the right to human review; most firms have not built the infrastructure to deliver it at scale.

“Which AI decisions is each member of your ExCo personally accountable for under SM&CR, and can they evidence oversight of those systems today?”

Market-level dimensions

Three dimensions sit above the firm's choices but shape what each firm can do.

The trust commons

Trust in financial services is shared. One firm's high-profile AI failure depletes trust across the industry. Each firm has an incentive to drift toward Default; the industry collectively pays when any one of them produces a public failure.

Vendor concentration

Most UK insurer AI runs on a small number of vendor platforms. If one vendor's model contains a flaw, multiple insurers exhibit it simultaneously. Systemic risk that does not appear in any individual firm's register.

Aggregator-mediated discovery

Most consumer insurance is bought via aggregators. The aggregator's AI is the most consequential AI in 'Becoming a customer', upstream of any insurer's system, invisible to the insurer, not directly overseen.

Conclusion

This discussion is not intended to prescribe a single “right” answer. The appropriate choices will differ by firm. What it aims to do is clearly articulate the decisions being made, often implicitly, so they can be addressed deliberately at the right level.

Some firms have arrived at their current position through gradual, operational drift. For them, the priority is recognising how these choices have accumulated and what they mean for the customer relationship. Others have made conscious decisions that align more closely with the Default state. For them, the challenge is whether those choices remain defensible in a changing regulatory and societal context.

Ultimately, the key question is not whether AI is being used, but how it is being governed in practice.

“Are you governing what your AI is actually doing, or what you think it is doing?”

This is where the real work begins.

For further information, please get in touch with your usual Crowe contact.