"Worth it" is a number, not a feeling

"Is this AI tool worth it?" has a one-number answer. How to run the minutes-per-pound review, the traps that flatter bad tools, and when to cancel without guilt.

Ask a founder whether their AI tools are worth the money and you'll get vibes: "the writing one's great", "I should use the other one more", "I forget what the third one does". Vibes are how subscriptions survive. No employee would keep their job on "I should use them more" — but tools escape the performance review entirely, because nobody defined what performance means.

The decoded idea: define it as one number — minutes back per pound. Estimate the minutes a tool honestly saved you last month, divide by what it costs, and you have a score you can rank the whole stack by. It's deliberately crude. Crude and consistent beats sophisticated and never-done, and this review takes ten minutes a month.

The practical move: the monthly scoring run. For each tool in your stack, three steps:

1. Estimate honestly. "The writing assistant drafts five client emails a week, saving maybe ten minutes each" → about 200 minutes a month. Where you're unsure, estimate low — optimism is how zombie tools survive.

2. Divide by cost. 200 minutes on a £15 plan is roughly 13 minutes per pound. Do it for everything, including the tools bundled inside other tools.

3. Rank and act. High scorers (10+ minutes per pound is a decent bar, but set your own) earn deeper use — more workflows, better briefing. Middling scores get one month of retraining: better prompts, custom instructions, a proper job description. Zeros get cancelled today, without a farewell tour.

One score deserves a footnote rather than a cancellation: the tool that saves few minutes but prevents expensive errors, or takes a dreaded task off your mind. Fewer errors counts. Calmer weeks count. Note that value in words next to the number — truthfully, not as a rescue story for a tool you're sentimentally attached to.

Slide from this week's carousel — the whole review in three steps.

The responsible-AI note (vet → disclose → verify). Keep the arithmetic honest in both directions. Time you spend checking a tool's output counts against its score, not for it — a drafting tool whose numbers you must triple-check may be a net negative wearing a productivity costume. Switching costs are real too: the learning curve of a replacement belongs in the maths before you migrate. And when a tool does get cancelled, close the loop properly — export your data, delete what the vendor holds, and revoke any access it had to your accounts. A cancelled subscription with a live API key is not cancelled. As ever with money-adjacent frameworks: this is an operating discipline, not financial advice.

Where founders get this wrong. Counting novelty as saving — the first week with any new tool feels fast; score it in month two, when the honeymoon has worn off. Scoring in December — an annual review means eleven months of paying for zeros; monthly is the cadence that catches drift. And letting the number decide alone — the score ranks candidates for attention; you still make the call, because you know context the arithmetic doesn't.

What this looks like after a month. The first run usually finds one obvious zero (cancel), one surprise star (use it deeper), and one awkward middle case (retrain, then re-score). The stack gets smaller and the remaining tools get better briefed — which raises their scores again. That's the quiet flywheel: measurement doesn't just trim the stack, it teaches you to run what survives properly.

This week's closer.

One decoded AI move, every week. Follow @aithebrief on Instagram for the daily version, or subscribe free on Substack for the deep one.