How to Choose an AI Tool Without Being Sold To

Every AI tool demos brilliantly. That is what demos are for. They are run on clean data, by someone who knows exactly which prompt works, in a scenario chosen because it works.

Your Tuesday will not look like that.

Start with the job, not the tool

The most expensive mistake is shopping before you have written down what you need. You end up buying a capable tool for a job you never clearly defined, and then defending the purchase.

Write one sentence: "I want to spend less time on ___, which currently takes ___ per week." If you cannot fill that in, you are not ready to look at tools.

Six questions a brochure will not answer

1. What is it genuinely good at? Not the feature list. The one job it does better than the alternatives.

2. What is it commonly oversold as doing? Ask the vendor directly. Good ones answer honestly; the answer tells you a lot either way.

3. What happens to my data? Is it used for training? Where is it stored? Can you delete it? If you handle client information, this is not optional.

4. Will my team actually use it? A tool that needs a champion to survive will die when that person is busy. Adoption is a feature.

5. What does leaving cost? Can you export your data in a usable form? How much configuration would you rebuild? What integrations would break?

6. Do I already pay for something that does this? Uncomfortably often, yes. Check your existing subscriptions before adding another.

Trial against your own work

Do not trial with the sample data. Take a real task — an awkward one, not your best case — and run it through. Then take a second one.

Two real tasks tell you more than a fortnight of poking at features.

Watch the total cost, not the sticker

Per-seat pricing looks small until you count the seats. Usage pricing looks generous until someone runs a batch job. And there is always the unpriced cost: the hours spent learning it, configuring it and explaining it to people.

A tool that costs more per month but takes an afternoon to adopt often beats the cheaper one that takes a fortnight.

Give it a review date

Put a date in the calendar three months out. On that date, answer one question: has this saved the time I bought it for?

If the answer is no, cancel it. The sunk cost is already sunk. What you are deciding is whether to spend the next three months' fee, and that decision has nothing to do with what you have already paid.