YouGov published its latest AI Index this week. Two numbers stood out to me.
93% of British adults have at least one worry about using AI. More than half, 54%, say they don't trust its answers or how accurate they are. And yet 51% use AI tools at least once a week. Even among people with a worry, half still use it weekly.
So the worry isn't stopping anyone. People are using it and doubting it at the same time.
I think that's the right instinct. It's how you'd treat a new starter. You're glad of the help, you check their work, and you don't give them the bank login on day one. The question for a small business isn't whether to trust AI. It's which jobs to give it first, and how you'll know when it's wrong.
Give it jobs you can check
The best first jobs for AI are questions you already have a way to answer. What sold yesterday. Which invoices are overdue. How many of a product are on the shelf. Which purchase orders haven't arrived.
You can check every one of those against your own systems. If the AI gets them right, you've learned something. If it gets them wrong, you've learned something more useful: where your data doesn't add up.
I made a similar point about using AI on the boring jobs. Boring jobs are where you can see whether it works.
Make it show where the answer came from
An answer with no source is a guess. A good AI tool for a business should be able to point at the thing it read: the order, the invoice, the stock record.
That matters most when your systems disagree. If Shopify says ten in stock and the warehouse says six, an AI that picks one without telling you which is worse than no AI at all. Decide first which system owns each number. I've written about that for Shopify and the warehouse and for Xero.
Keep a person on anything that changes something
Reading is low risk. Changing things isn't. Stock adjustments, price changes, invoices, payments, supplier emails. For those, AI should draft and a person should confirm.
Xero has built its new Gemini app this way. Its announcement says any action that changes financial data, like issuing an invoice or sending a payment reminder, needs the user to confirm it first. That's a sensible default for any business.
Write it down, too. A short list of what the AI can read, what it can draft and what it can never touch. I covered that in write down what your AI can touch.
Why AI gets business answers wrong
Most wrong answers aren't the AI being stupid. They come from the AI not knowing how your business works. It doesn't know that "in stock" for you means on the shelf and not on a lorry. It doesn't know which of three spreadsheets is the real one. So it guesses.
That's the problem we're working on with Bud, which is in beta. It holds how your business works, your systems, your words, who owns which number, so the AI tools you use start from that and not from a guess.
Where to start this week
- Pick three questions you ask every week and already know how to answer.
- Ask an AI tool connected to your systems the same questions, and compare.
- Write down what it can read, what it can draft and what it must never change.
- Keep a person on anything that touches stock, money or customers.
If your answers live in three systems that don't agree, fix that first. Talk to us if you want help.
Sources
- YouGov, Britons are concerned about AI, but half still use it every week, 27 Oct 2026: yougov.com/en-gb/articles/55825-britons-are-concerned-about-ai-but-half-still-use-it-every-week
- Xero media release, Xero brings real-time financial intelligence to Google Gemini, 14 Oct 2026: xero.com/us/media-releases/xero-brings-real-time-financial-intelligence-to-google-gemini/