Where to start with AI when the board is already asking

This is the most common conversation I have, and it opens the same way almost every time. Somebody says they know they should be doing something with AI and they have no idea what. They are usually a little embarrassed about saying it out loud.

They should not be. The embarrassment is manufactured. There is an enormous volume of noise pointed at business owners right now, from vendors, conferences, and competitors announcing partnerships that will not survive the year, and almost none of it tells you what to do on Monday morning.

Do not start from the technology

The wrong first question is which AI product should we buy. It leads to a demonstration, the demonstration is good, and you end up owning a solution that has to go looking for a problem. That search is usually conducted by whoever signed for it, which makes it a poor search.

The better first question is duller. Where in this business does somebody make the same judgement over and over again, where being wrong costs real money?

The shape of the answer is usually something like: whether to extend credit to a new customer, which enquiries are worth chasing this week, whether a shipment is going to be late while there is still time to tell the client, which invoices are wrong before they go out, which of two hundred CVs are worth reading. What those have in common is that the decision happens often, a person currently makes it using judgement, and the cost of getting it wrong is a number somebody in finance could put on a page.

Then check whether you already hold the information

For whatever you picked, ask a single question. Do we have a record of the last two years of these decisions and how they turned out?

If yes, you have something to work with, and the project is real. If the honest answer is that it lives in people's heads, or in email, or in a folder of scanned documents nobody has ever read as a set, then your first project is smaller and duller than you hoped, and it starts with writing that down.

That is not a failure. It is the cheapest thing you will ever find out. Two weeks to discover it costs you a fortnight. Five months to discover it costs you the programme.

Pick something small enough to finish

The pressure will be to pick something visible, because a small win is harder to present than a big ambition. The instinct to make it large is exactly backwards.

Take the smallest version that would still be worth doing on its own. One product line. One branch. One type of document. Get it into real use, let people work with it for a month, and then decide what happens next with information rather than with hope.

You learn more from one finished small thing than from six months of a large one, and you get something the organisation has actually seen work. The second project is far easier to fund when the first one is running in front of people.

Whatever you pick, agree what it costs you today, in a form your finance director would accept. Hours. Dirhams. Errors a month. Days of delay. Measure it now, while the old way is still running, because in three months there will be nothing to compare against.

This is the single most skipped step I see, and it is why so many organisations genuinely cannot tell you whether the last thing they bought worked.

Be willing to decide this is not the year

I say this as somebody who sells this work. Sometimes the right answer is to spend the money on getting your records in order and come back to it in twelve months.

If your stock figures are unreliable, your customer list exists in three versions, and nobody can say what last quarter's margin was by product line, then an AI project will produce confident answers built on all of that. The confidence is the dangerous part. A person reading a bad spreadsheet knows to be suspicious. A clean-looking answer on a screen gets believed.

Nobody is behind in the way the marketing implies. The organisations that do well with this over the next few years will be the ones whose records are clean and whose process is written down, and both of those are available to you now without buying anything from anybody.

Something you can do this week

Write down three decisions your people make repeatedly where a mistake is expensive. Against each one: roughly how often it happens, roughly what a wrong one costs, and whether you have a record of the past ones. Half a page, no software involved.

That half page is a better starting point than any vendor meeting, and it is what I would ask you for anyway. If you want a second opinion on which of the three to take first, send it to me and I will tell you what I think, including if the answer is none of them.