A contract clause, policy restriction or warranty can change whether a commercial policy meets a client's needs. For Matthew Mawson (pictured), managing director of Sheffield-based Eximia Broking, those details are exactly where automation should stop and a broker's advice should begin. His clients in road transport, construction and plant hire, he said, need more than an online question set to get it right.
"Back office and presentations should be automated as much as possible," Mawson said. "However, client-facing, advising and risk evaluation should be carried out by the broker." He said understanding commercial risks requires time spent with the client and their business, something that cannot be "undertaken by AI." Without that understanding, brokers cannot identify the most suitable cover.
The British Insurance Brokers' Association and Markel have taken a similarly measured line since 2025, when their joint AI guide warned against treating the technology as a fix for everything. A caution that has since grown into the Broker AI Academy, whose founding cohort of 20 brokers completed its first two-day session in Leeds earlier this month.
The same distinction is playing out across the market. Some brokers argue that personal relationships are becoming more valuable as AI takes over routine work. "You might as well be a comparison site," said Charlie Hicks, managing director of EIC Insurance Services. "If you're doing that, you've added nothing to the relationship."
For clients in road transport, construction and plant hire, Mawson said the gap between a portal and a broker shows up in contract detail. A portal cannot provide the depth and knowledge required to properly insure all the risks, whereas a broker with the right experience knows which questions to ask. That includes understanding whether a job falls under the Joint Contracts Tribunal (JCT) suite used across UK construction, the Construction Plant-hire Association (CPA) model conditions for hired plant, or the Convention on the Contract for the International Carriage of Goods by Road (CMR), which has carried the force of law in the UK since the Carriage of Goods by Road Act 1965. Mawson said brokers also need to flag what cover uplifts are required, which policy restrictions should be negotiated out and which warranties a client needs to be aware of.
"I've seen numerous examples of policies bought online that are not fit for purpose," he said. Question sets are often ambiguous, clients don't understand the consequences of a wrong answer, and once a policy is bought, the document is rarely read properly until something happens, by which point it is too late.
Other senior figures describe the same problem from the insurer side. Speaking at AIRMIC 2026, Stuart Sutherland, head of UK retail at Liberty Specialty Markets, described a furniture retailer whose policy was written as if the business were still a simple shopfront, when it had grown to design, manufacture, deliver and install its own products – several distinct exposures sitting undetected under one programme. "Back to basics means asking: tell me about your business," he said. "What keeps you up at night? Before we talk about coverage."
Mawson does not think a people-first model is necessarily slower than an automated one, provided the right information and a clear understanding of risk and objectives are established from the outset. He said higher acquisition costs can squeeze margins, although stronger renewal retention should help offset them.
"A fully automated process works for scaling low-value products that churn. People-first helps retention, thus smoothing margin, and provides recommendations that help scale a business."
The starting point is whether the client needs advice and how complex their business and insurance requirements are, Mawson said. Automation can smooth the client journey, “but don't leave the advice to an AI data centre.” Those needs should determine where automation stops and a broker steps in.
