It’s the most common opening line I hear from business owners: “We know we should be doing something with AI.”

I understand exactly where it comes from. Every supplier email, every LinkedIn feed, every industry event is carrying the same message: your competitors are doing this, you’re falling behind, buy now. Under that pressure, “we should do something” feels like responsible leadership.

But as a brief, it’s unbuildable. “Something with AI” has no owner, no number, no finish line. Projects born from it follow the same arc every time: a tool gets bought, a pilot gets announced, enthusiasm lasts six weeks, and a year later the subscription is still billing while nobody remembers the login. The conclusion drawn is “AI doesn’t work for us” — when what actually failed was the brief.

The one-sentence upgrade

The difference between the projects that die and the ones that pay for themselves is embarrassingly simple. The good ones start with a sentence shaped like this:

“[Specific problem] is costing us roughly [number], and in six months success looks like [measurable outcome].”

“Manual reporting eats about eight hours a week — call it £9,000 a year — and in six months I want those numbers arriving automatically.” “We don’t find out customers have gone quiet until they’re gone; in six months I want a weekly list of accounts going cold.” “Quoting takes three days and we lose jobs to faster competitors; in six months I want same-day quotes.”

Notice what that sentence forces. A specific problem — which means you had to choose, which is the actual hard part. A number — which makes ROI a calculation instead of a vibe, and tells you what a sensible budget is. A measurable outcome — which means in six months you’ll know whether it worked, instead of debating it.

If you can’t fill in that sentence yet, you don’t have an AI problem. You have a prioritisation problem — and buying a tool won’t solve it, it’ll just give the vagueness a monthly cost.

Two traps on the way to the sentence

Starting with the shiniest thing. Forecasting and “predictive” anything is the most seductive offer in the market right now. But prediction built on numbers nobody trusts is expensive astrology. If the earlier pieces in this series flagged gaps in your data foundations, fix those first — prediction is a phase-two project, and the businesses that treat it that way are the ones it actually works for.

Mistaking a mood for a mandate. “The team is excited about AI” is lovely and worth nothing without one named person who has actual time and actual authority to own the project. In most small businesses that defaults to the owner, “in the hours that don’t exist” — which is how six-week enthusiasm becomes month-eleven shelfware. Before you start anything, answer honestly: whose diary does this live in?

Where this series has been heading

Over five weeks I’ve walked through the five things that decide whether a business benefits from data and AI: how decisions get made, whether the data foundations exist, who’s already using AI and how openly, whether anyone’s decided the rules, and — today — whether you know which problem to fix first.

Not one of them is about which AI tool to buy. That’s the point. Readiness isn’t AI usage. A business with clean data, clear priorities and no AI at all is better positioned than one with ChatGPT everywhere and chaos underneath.

Later this week I’m launching something that turns those five dimensions into a proper diagnostic — a way to see, in about eight minutes, exactly where your business stands and which gap to close first. More on that shortly.

In the meantime: can you fill in the sentence? What’s costing you how much — and what does fixed look like in six months?

Delphi Decide helps owners turn “we should do something with AI” into one sentence they can actually build. If you’d like help filling yours in — get in touch.