Where AI is actually paying its way in UK SMEs

Most UK business owners have tried AI by now. Far fewer can point to a job it does for them every week. That gap is where the money is.

Broad, but shallow

According to the Office for National Statistics, 35% of UK firms with 10 or more staff now use AI, up from 12% in late 2023. But only around 1 in 10 of those firms use it extensively. Over half of UK employees (55%) say they use AI for work or study.

Use is also very uneven. In information and communications, 58% of firms use AI. In construction, it's 13%. And the number one barrier firms give isn't cost or skills. It's “we can't work out where it fits”.

In other words, adoption has nearly tripled and depth has barely moved. Plenty of people have a gym membership. Not many are going.

You don't have an AI problem

When we sit down with an owner, the issue is rarely the technology. It's visibility. Nobody can see the whole job from start to finish, so nobody can see where the hours go. Your team and your customers hear the squeaky door every day. You stopped hearing it years ago.

That's why we always say: work first, tool second. A drop of oil fixes a squeaky door. A smart home system doesn't.

The six jobs AI is good at

Most useful business AI does one of six jobs, and each one matches a kind of drag you'll find in almost any business:

• Typing it twice → read and extract. “Turn that email into an order.”
• Chasing → remind and follow up. Nudges that never forget.
• Hunting → search your own files. “What did we quote them last time?”
• Checking by hand → compare and flag. Only show me the ones that don't match.
• Slow to respond → draft and reply. A first answer in minutes, approved by a person.
• Only one person knows → capture and share. What one expert knows, available to everyone.

What it looks like in different businesses

A manufacturer might use it to read supplier invoices and match them to purchase orders, flagging only the ones that don't agree. A construction firm might use it to summarise tenders and draft site reports. A professional services firm might use it to draft letters and pull answers from its own past work. A retailer might use it to plan stock and rotas from demand. The six jobs stay the same. The paperwork changes.

Why AI projects stall

We see the same four mistakes again and again:

1. Automating a mess. Agree one way of doing the job first. A faster mess is still a mess.
2. Starting too big. One job, one owner, one number. Not “AI for customer service”.
3. Building for the easy 80%. Ask your team what goes wrong. That list is your design brief.
4. No way back to a person. When the system isn't sure, it should hand over, not guess.

Where to start

Ask your team one question tomorrow: “What's the most annoying thing you do every week?” Then map that one job, step by step, and look for the six signs of drag. Our AI opportunity workbook walks you through it, and the Earns Its Keep test helps you score what you find.