What happened

Gene Marks’ Guardian column on 30 August 2026 considered lessons smaller businesses could take from corporate AI adoption.

Here is my proposed way to turn that question into a manageable decision: design a small trial around one recurring problem, then judge the result against the work already being done.

A modest business should be able to learn something useful without committing its entire operating model to an experiment.

Why it matters

For a small team, the cost of a failed experiment is often time the same people needed for customers. A project that requires constant attention from the owner can become expensive even when the software subscription is cheap.

Begin with a task the team understands well. It should occur often enough to evaluate, have a recognisable good result and permit human review before anything consequential happens.

Consider an illustrative independent hotel preparing answers to common guest enquiries. A suitable first trial might draft responses from an approved information sheet. Staff would still check and send each reply. The trial would exclude payments, identity documents and decisions about exceptions.

This is a proposal, not a reported case study. Its purpose is to show how a broad interest in AI can become a defined piece of work with a clear boundary.

The bigger shift

Write a one-page trial brief before choosing the tool. State the problem, the current process, the permitted information and the person responsible. Add the expected improvement and the conditions that would stop the experiment.

For the hotel example, collect a sample of ordinary enquiries with personal details removed. Ask staff to assess whether each draft is accurate, appropriate and useful. Record corrections and the total time required, including preparation and checking.

Include awkward questions: a request for a guarantee the hotel cannot offer, an outdated facility description and an enquiry that requires a manager’s judgement. The system should recognise the limits of its approved information.

At the review meeting, compare accepted responses and staff effort with the existing method. Discuss whether maintaining the information sheet is realistic. A successful demonstration on prepared examples is only the beginning of that assessment.

The wider AI strategy framework can help connect the next step to business priorities, even when the organisation is small.

My take

I would give the first trial a narrow ambition: produce enough evidence to make the next decision easier.

That decision might be to continue, revise the process, try another approach or stop. All four can be useful outcomes if the team understands what it learned. An experiment becomes wasteful when it continues without a clear question or review point.

Keep the employee doing the work involved throughout. They will often notice whether the system removes a frustration or creates another checking task. Give their observations a place alongside the numbers.

Finally, write down what must remain true for the trial to stay useful: accurate source information, available review time and an acceptable error pattern. Revisit those conditions before extending the system to another task.

A small business does not need an elaborate AI programme to begin. It needs one worthwhile problem, a controlled test and the discipline to act on what the evidence shows.