I used AI badly so you don't have to

Confession time: I use AI in this business. It helps draft documents, summarise long reports and tidy up my writing — including some of this website.

Second confession: I haven't always used it well. Early on I had misunderstood the scope of a project, and wasted time and was behind, I used ai to gather some information and skipped the checking I'd apply to any other source, and produced something that did not read well. It was awful. Nobody was harmed — but my pride has a corrective action raised against it and I decided to void the invoice for the whole job - not ideal.

That's the real risk of AI at work. Not the robot uprising — mediocre work, nicely formatted, with your name on it.

So: four rules, learned the hard way, free to borrow.

• Confidentiality first. Client information never goes into a tool unless you know exactly what it does with the data. Can't answer that? Then the answer is no.

• The human signs the work. AI can analyse data, start drafts, prepare summaries, even review for typos and errors but it doesn't decide. Its mistakes arrive well-formatted and plausible, which makes them far easier to miss than human ones. If you're not willing to stand behind it, it doesn't go out.

• Be transparent where it matters. Nobody needs a disclaimer on every email — but clients paying for expertise shouldn't be receiving unchecked autocomplete. Pass on the savings, if you can be more efficient using ai, then the client should also benefit,

• Keep your skills. If the tool vanished tomorrow, you should still be able to do the job. AI should make good people faster, not businesses helpless.

This is just risk management - my favorite topic

Work out what could go wrong — a privacy breach, a confident wrong answer, a staff member freelancing on an unapproved app — assess it on evidence rather than hype, and control it in proportion. For most small businesses that's a one-page AI policy: approved tools, what data goes in, who checks what comes out.

Tenders already ask suppliers about data handling; AI questions are close behind. You can answer with a sensible page you wrote calmly or write one in a panic the night before the tender closes. I've tried calm. Recommend it.

Want an AI policy that fits your management system — and on one page? Get in touch.