If you run a small or mid-sized business, you’re getting the AI message from every direction. Your software vendors, your competitors’ LinkedIn posts, the trade press, probably your own team. Some of it is real. A lot of it is noise. The hard part isn’t finding things to try, it’s working out which one is worth your limited time and money first.

We work with businesses in exactly this position, and the ones who get value from AI tend to do the same few things. The ones who waste money tend to make the same few mistakes. Here’s both.

Start with a problem, not a tool

The most common way to waste money is to buy AI and then go looking for something to point it at. It should be the other way round. Pick a real problem that costs you time or money today, one you can describe in a sentence to someone who doesn’t work for you. “We spend two days a month copying figures between systems.” “Our quotes take too long because the information is scattered.” Now you’ve got something to measure a solution against.

If you can’t name the problem clearly, you’re not ready to spend on the solution.

Do the cheap experiment first

For most first projects there’s an off-the-shelf tool you can trial for the price of a subscription. Use it. Give it to the people who’d actually use it, for a real task, for a couple of weeks. You’ll learn more from that than from any amount of research, and it costs almost nothing.

Sometimes the cheap tool solves the problem outright and you’re done. Sometimes it gets you eighty percent of the way and shows you exactly where the last twenty percent, the part specific to your business, would need something built. Either way you’ve spent very little to find out.

Know when to stop renting and start owning

There’s a point where off-the-shelf stops making sense. Usually it’s one of three things: the per-seat cost has crept up as more people use it, the tool almost fits but the gap is now costing real time, or the thing you need depends on your own data and processes that no generic tool understands. That’s when a custom build starts to pay back, and not before.

Plenty of businesses build too early, before they understand the problem. Plenty build too late, after years of paying to work around a tool that never quite fit. The judgement is all in the timing, and it’s worth getting a straight opinion before you commit either way.

The mistakes that cost the most

Buying out of fear of missing out. If the honest reason is “everyone’s doing it,” stop. That’s how tools end up bought, unused, and quietly cancelled a year later.

Ignoring your data. Most of the interesting things AI can do for a specific business depend on that business’s own data being usable. If yours isn’t, that’s the first project, and it’s a valuable one regardless.

Expecting it to think for you. AI is good at handling volume, spotting patterns, and doing the repetitive middle of a task. It’s not good at judgement, and betting on it for decisions that need real expertise is where trust gets lost.

A sensible first step

If you want a shortcut through all of this: write down the one problem that costs you the most time each month, and ask whether the value is in the task itself or in your data. Task alone, try an off-the-shelf tool this week. Data, that’s worth a proper conversation about building something.

We do that kind of scoping with businesses all the time, and we’ll tell you honestly when the answer is “you don’t need us for this.”