What AI can actually do for a five-person business

There is a gap between what AI is said to do for businesses and what a five-person company can actually switch on this month. Most of the loud examples come from organisations with a data team, and the advice does not survive the translation downwards.

Here is the version that does.

What actually works at small scale

Drafting. First versions of anything written — a service description, a reply to a recurring question, a quote covering letter. Editing beats staring at an empty page, and this is where nearly all the real time saving is.

Summarising and extracting. A long email thread into three decisions. A supplier's PDF into the four numbers you needed. Reliable, unglamorous, useful weekly.

Translating. Genuinely good now, and it opens a second language on a website for the cost of reading it through.

Explaining things you half-know. A regulation, an error message, a contract clause. Not as a final answer, but as a way to ask a better question of the person who does know.

What does not work yet at small scale

Anything where being wrong is expensive and nobody checks. Prices, availability, legal wording, medical or financial specifics. The failure mode is confident and fluent, which is the worst kind.

Agents left to act on your systems. The demonstrations are real and the production versions need monitoring, permissions and someone accountable — all things a five-person business does not have spare.

Anything requiring your judgement about your own business. Which customers to chase, what to stop selling, what your prices should be. It will produce an answer. It has no way of knowing whether it is right.

The test before you adopt anything

Three questions, in order:

1. How often do I do this? Weekly or more, or it is not worth setting up.
2. What happens when it is wrong? If the answer is "a customer is misled", a human checks every time or you do not use it there.
3. Can I tell when it is wrong? This is the one people skip. If you cannot evaluate the output, you are not using a tool, you are trusting a stranger.

Anything passing all three is worth adopting today. Anything failing the third is worth avoiding regardless of how impressive the demonstration was.

About the statistics

You will read that some enormous percentage of companies are "using AI" and a tiny percentage have it in production. Treat those numbers the way you would treat any vendor statistic about a market that vendor sells into — including when the vendor is us.

What matters for your business is not the average. It is whether the three questions above give you a yes.

Pick one task, this week

Not a strategy. One task you do weekly, that involves writing or summarising, where you would notice a bad result immediately.

Do it both ways for a fortnight. Keep it if it saved time, drop it if it did not, and ignore everything you read in the meantime.

Related: for websites specifically, the limits are more concrete, and the copy question has its own answer.

Frequently asked

Is AI worth it for a business with five people?

For drafting and summarising, almost always — those are weekly tasks where a bad result is obvious immediately. For anything customer-facing and factual, only with a person checking every time, because the failure mode is fluent and confident.

Should I worry about AI getting facts wrong on my website?

Yes, for anything a customer can hold you to: prices, guarantees, delivery times, availability. Generated text is not more likely to be wrong than a tired human, but it is far more likely to be wrong in a way that reads as authoritative.