Starting · Briefing

Why the first AI tool you tried was impressive and useless

Most owners we speak to have tried one AI tool. It wrote something clever in seconds. They were impressed, then they tried to use it for real work and it did not fit. They drew a conclusion: it is a toy, or it is not for a business like theirs.

The conclusion is understandable and mostly wrong. The tool was fine. The way it was used was the problem.

A demo is a general answer to a general question

When you ask a tool to write a paragraph about your industry, it answers from everything it has seen. It does not know your customers, your prices, your suppliers, your tone or the arrangement you made with your best client in 2019. So it produces something plausible and generic. That is impressive for about ten minutes.

Real work is the opposite. It is specific, and it depends on facts that live in your head, your inbox and a spreadsheet nobody else understands.

Why it stopped at "impressive"

There are usually four reasons.

• It was used for a task with no clear finish line, so nobody could say whether the result was good.
• It was never given your information, so it guessed.
• It lived in a browser tab, away from the place the work actually happens.
• Nobody owned the outcome. It was one person's experiment, and it stopped when their week got busy.

None of these is about the technology. They are about fit.

What useful looks like

A useful tool is boring to describe. It does one repeated job, for one person, with your information, inside the system that person already uses. For example: when an enquiry arrives, it reads it, finds the matching customer record, drafts a reply in your voice and leaves it for you to check.

That is not a clever demo. It is a small piece of plumbing. It saves twenty minutes, forty times a month, and nobody applauds.

The useful version is rarely the impressive one.

How to try again, properly

1. Pick one job that repeats and that you already do the same way each time.
2. Write down what a good result looks like, in a few sentences, as if for a new employee.
3. Decide who checks the output, and how quickly.
4. Give the tool the information it needs for that job, and nothing else.
5. Run it beside your normal process for two weeks. Compare.

If it does not hold up, you have learned something specific about that job. That is worth more than a general opinion about the technology.

When it is not worth doing

Some jobs should not be handed over. Anything where one wrong answer is costly and hard to undo, anything that turns on knowing the person, anything you cannot describe. We will tell you when this is the case, including when the honest answer is that you do not need us.

The short version

The first tool was a demonstration. The second attempt should be a small, specific job with an owner and a check. Done that way, the same technology usually turns out to be useful.

If you would like help picking the job, tell us what is going wrong and a person will reply.