Metered AI killed the subscription question. Work out what one finished proposal, report or answered enquiry actually costs — in about ten minutes, no spreadsheet.

What is cost per outcome, and why replace cost per seat?

Cost per outcome is what one finished thing costs you: one client report, one drafted proposal, one answered enquiry, one month of scheduled social posts. Not what the tool costs. What the result costs.

Cost per seat is now close to meaningless, for a simple reason: the seat stopped being the thing you buy. When AI billed at a flat £20 a month, price per seat told you everything, because usage was free at the margin. Now that a growing share of AI work bills per token or per credit, two businesses on the identical plan can pay very different amounts — and the difference is not how many people you have. It's how efficiently you ask.

There's a second reason the seat number is fading. Flat prices are not staying flat either: Microsoft raised 365 prices on 1 July 2026. So the subscription line moves, the usage line moves, and the only number that stays comparable month to month is the one attached to a unit of work you actually sell.

How do I work it out without building a spreadsheet?

Take one repeatable job you do often. Not the whole business — one job. Then four lines:

1. Name the outcome precisely. "A finished 800-word client update, ready to send." Vague outcomes give vague numbers.
2. Count the runs it really takes. Be honest. Most people say one. Most people do three: a first attempt, a re-brief, and a tidy-up.
3. Multiply by the run cost. Your dashboard shows usage; divide last month's charge by last month's jobs if you can't see per-run figures. On Meta's Business Agent, for instance, an AI-handled conversation sits around 4–5 cents at $2 per million tokens — so the unit is small, and the multiplier is what matters.
4. Add the minutes you spent. At your hourly rate. This is usually the biggest number on the list by an order of magnitude, and leaving it out is how people conclude that AI is expensive when their own time is the expensive part.

Now you have a real figure. Most solo founders find the AI cost is trivial and the re-run cost — in time, not tokens — is where the money goes. That's a useful discovery, because re-runs are the part you control.

Do it for one job and you have a number. Do it for three or four and you have something better: a ranking. The jobs sort themselves into ones where AI is doing genuine heavy lifting and ones where you've added a step to a process that was already fine. The second category is more common than anyone admits, and it never shows up in a subscription cost.

What does this change about how I prompt?

It turns sloppy briefing from a style preference into a line item. Every re-run is billed. Three vague attempts cost three times what one clear brief costs, and the clear brief usually produces better work anyway — so the incentive and the quality now point the same way for the first time.

Practically, that means front-loading. Say who it's for, what good looks like, what to leave out, and how long it should be. Ask for the exact format you need rather than reformatting afterwards. And edit the answer instead of re-running it, because "try again" throws away the parts that were fine and re-bills you for them.

Before any long job, one line saves two runs: "Before writing, give me your plan in five bullets, then wait for my go-ahead." That single habit is worth a post of its own — getting a quote before the work starts is the cheapest correction you'll ever make, because it costs a few hundred tokens to catch a misunderstanding that would otherwise cost a full run.

What do you do with the number once you have it?

Three things, and the third is the one people skip.

Compare it to the alternative, honestly. If a finished client update costs you £2 of tokens and 15 minutes, against 90 minutes before, that's the real comparison. Not against zero.

Price from it. If AI made a deliverable four times cheaper to produce and your price hasn't moved, you've quietly given the saving away. That was already true when the speed arrived; metering just makes it measurable. AI made you faster — did your prices notice? is the uncomfortable half of this week's theme.

Find the outlier. One job in your list will cost far more per outcome than the rest. That's your automation target — not the job you find most annoying, which is the usual selection method and a much worse one. You can't spot the outlier without the numbers, which is the entire argument for spending ten minutes on this.

Do one job this week. Not all of them. The habit is the deliverable — a number you can check next month against a number you wrote down this month. Everything else in a metered world is guessing with extra steps, and the whole point of a meter is that you no longer have to guess.

Not financial advice — vendor rates vary and change; verify your own usage and pricing before making commercial decisions.

For the prompt set behind these briefs — free, 5-min read: 7 AI Prompts That Run a One-Person Business.

Related Decodes

• AI Just Became a Utility Bill — the week's hub: what metered pricing changes, and the 5-minute bill audit.
• Your Week Leaks 10 Hours — find the jobs worth costing in the first place.
• Automation Isn't Free. It's Just Billed Later. — the judgement version of the same sum.

AI the Brief (@aithebrief) — AI advice you can act on and trust, for time-starved founders.

TL;DR: Metered AI pricing makes cost per seat close to meaningless, because two businesses on the same plan now pay different amounts depending on how efficiently they ask. Replace it with cost per outcome: name one repeatable job, count how many runs it honestly takes, multiply by the run cost, and add your own minutes at your hourly rate. Most founders find the tokens are trivial and the re-runs are expensive — which makes clear up-front briefing a cost control, not a style preference, since every re-run is billed.