Will AI replace my insurance broker?

There’s a lot of noise about AI taking jobs, and brokers are on most of the lists.

Insurance is changing, but not in the way the headlines suggest. So, can AI replace your broker? No, not yet. For most businesses, the more useful question is what new risks AI is creating for them.

What is AI changing in insurance?

The industry is adopting AI fast, and governance is struggling to keep up. A June 2026 Earnix survey found 98% of UK insurers are using or planning to use generative AI, but only 28% of UK insurance leaders strongly agree their AI governance is keeping pace.

Claims is where AI is making most of a difference. It has made it easy to fake the evidence behind a fraudulent claim: photos of injuries or vehicle damage, garage reports, medical reports.

The numbers are heading the same way. The ABI reports insurers detected £1.34bn of fraudulent claims in 2025, up 14% on 2024, even though the number of cases fell. Fewer fraudulent claims, but bigger ones.

Insurers are on the back foot, and a lot of time and effort is going into what are effectively AI bouncers, tools that pick up on faked evidence that humans can’t spot.

This affects you even if you’d never make a false claim. The ABI estimates fraud adds around £50 a year to the average policyholder’s premium. And as insurers tighten their checks, honest claims get more scrutiny too.

Customers mostly see the gentler side, though: chatbots, quicker quotes, faster underwriting. It will be a few years yet before AI changes anything dramatically on the customer-facing side.

Why AI won’t replace your broker

I’m biased, but I think AI makes brokers more valuable, not less. There are three reasons.

People want to be understood. However good AI gets at generating quotes and explaining the finer details of a policy wording, most people still want someone who can explain it in a way they understand, and with some empathy when the news is bad.

Regulation needs a human answer. You can generate quotes with AI all day. But when a data model spikes the price, or declines cover outright, the customer doesn’t understand why. The FCA expects firms to be clear about why a price has changed and what a client needs to be aware of. That’s where the governance gap shows. Most insurers’ compliance frameworks aren’t ready for it, and AI can’t handle that conversation on its own.

Someone has to be on your side. An AI quote engine works for whoever built it. When a claim gets difficult, you need someone who knows the wording, knows the insurer and will push back on your behalf. That’s the job. It’s also why we share our communications with the insurer with our clients during a claim, so they can see us doing it.

The risk most businesses underestimate

Cyber and data is the risk most businesses underestimate, but probably not in the way you’d expect. There are two sides to it.

Phishing is getting harder to spot. In 2024, the National Cyber Security Centre warned that generative AI removes the spelling and grammar mistakes that used to give phishing emails away. That matches what we see. The badly spelled email from a strange address is going. Now you receive emails that are almost indistinguishable from legitimate ones, from senders that look legitimate. Phishing is still the biggest threat UK businesses face: the government’s 2025/26 Cyber Security Breaches Survey found it was the most common type of attack, hitting 38% of businesses.

The costly version is often simple. An email that looks like it’s from a supplier asks you to update their bank details. There used to be tells. Now there often aren’t. Before changing any payment details, phone the supplier on a number you already hold. It’s still the best defence.

The second risk is more innocent. We see it day in, day out, with our own clients and the people we talk to. For financial services and consultancy firms especially, AI is an attractive shortcut. So someone uploads a client’s policy schedule, or a contract they’re under NDA for, to ChatGPT, Gemini or Claude to ask it a question.

“Something an employee thinks they’ve done innocently immediately turns into a breach of an SLA or a contract. And that sensitive data has now been given to a third party.”

This isn’t rare. A 2025 study found 48% of employees had uploaded sensitive company or customer information into public AI tools. If you work under NDAs or data processing agreements, a single upload can breach them.

Where I’d start

If a client came to me worried about AI risk, my first advice would be to buy a cyber and data policy if they haven’t already got one.

Across our own book, 240+ clients hold cyber and data cover with us. Where it’s bought as a standalone policy, a typical premium is around £500. Prices range from £100 to £10,000+, depending on the size of the business and what it handles.

Uptake is highest in software: 78% of our healthtech clients, 75% in edtech, 71% in AI and machine learning, and 64% in SaaS. It’s lowest in more physical sectors, such as food and beverage (13%) and manufacturing (23%).

The figure that stands out to me is professional services and consultancy, at 20%. That’s the group most likely to be uploading client contracts and schedules into AI tools. These numbers only show cover placed through us, so plenty of those clients may be insured elsewhere. But if that’s your sector, it’s worth checking.

Cyber policies aren’t all the same. When you review one, check:

• Fake payment requests: whether losses from social engineering, such as a fake bank-detail change, are covered, and at what limit. This is often an optional extension with a lower limit.
• Breach response: whether it includes IT forensics, legal advice and the cost of notifying customers and the ICO.
• Staff mistakes: whether it covers a breach caused by your own team’s error, not only an outside attack.

The second is a change in mindset:

“Stop treating AI as a tech issue, and start treating it as a corporate governance issue.”

Most businesses aren’t there yet. The same government survey found 31% of businesses are using, adopting or considering AI, but only 24% of those have cyber security measures in place to manage the risks.

If your team uses public AI tools, the job is to put the right framework in place so they can use them safely. In practice, that comes down to four decisions:

• Approved tools: which AI tools are allowed, ideally business accounts that don’t train on your data.
• What never goes in: client data, contracts under NDA, personal data.
• Who checks output: who reviews AI-generated work before it reaches a client.
• Who owns it: who is responsible for the policy at leadership level.

We’re no different. We use AI to make our lives easier, raise our productivity and do a better job for our clients. The most important thing we’ve done, though, is build in human-in-the-loop architecture.

“You can use AI all you want to sense-check and give you a preliminary draft or answer. But it can’t leave the building. It can’t go to the customer until a human has seen it, looked over it and sense-checked it.”

So, can AI replace your broker? It can make a good broker faster. It can’t take responsibility for the advice, and it won’t fight your corner when a claim gets difficult.