George Robbins, head of UK markets at Shift Technology, says that before insurers ask AI to do more, they should make sure the process deserves to be automated in the first place
Everyone in insurance is talking about artificial intelligence (AI) and rightfully so. Claims volumes are rising, customer expectations are higher than ever and teams are under pressure to do more with less. But there is one mistake I see too often. Insurers try to automate a process before they have properly tested whether that process is fit for purpose.
AI is not a repair kit for broken workflows. If a claims process is too fragmented, too inconsistent or open-ended when it should have a closed feedback loop, automating it will not make it better. It will simply make the same weaknesses move faster.
That matters because insurance is not a sector where speed alone creates value. A faster bad decision is still a bad decision. In claims, fraud, underwriting and customer service, the real question is not “can AI do this?” but “should this step be automated and what still needs human judgement”.
Best uses for AI
The best uses of AI are not necessarily the most flashy ones. They are the ones that make life easier for experienced claims handlers, fraud investigators and underwriters.
For example, in claims handling, AI can triage incoming information, surface missing documents, flag inconsistencies and prepare the next best action. That is useful because it reduces admin, but more importantly because it gives the handler a clearer view of the case.
This is where AI creates value in insurance – not by replacing expertise, but by helping professionals apply it better. The machine can handle repetitive tasks, routine checks, surface options and help plan. The human can focus on the exception, the nuance and the decision that affects the customer. In a good process, that balance makes the outcome faster, fairer and more consistent.
This is also why governance cannot be an afterthought. Insurance is a regulated business. If the model is uncertain, if the data is incomplete, or if the case is sensitive, the system should know when to stop and hand it over to a person. Clear boundaries, clear escalation paths and clear accountability are not optional, it is what makes AI usable in a sector built on trust.
Too many AI projects fail because organisations start with the technology rather than the process. They ask how fast the system can work before they ask what the process is meant to achieve, who owns it and what a good outcome looks like for the policyholder as well as the business. In insurance, those questions matter – a claims decision affects a customer. A fraud decision affects trust. An underwriting decision affects risk.
If AI is applied carefully, it can improve consistency, reduce administrative burden and help teams respond more quickly to customers. If it is applied carelessly, it can implement inefficiency at scale or worse still, make decisions that break trust.
So before insurers ask AI to do more, they should make sure the process deserves to be automated in the first place. The goal is not to automate bad habits, but to build processes that are faster, fairer and more reliable.
Find out more about Shift Technology here
