Fictional company (Kestrel Supply Co.). SYNTHETIC data. ILLUSTRATIVE prices. Proof of concept.
Decision memo: which AI tools to keep
Data as of 2026-06-29. Every money figure is for one month. Prices are ILLUSTRATIVE.
Activity is not value. This method can show what is not used. It cannot prove what is worth keeping.
The answer for each tool
| Tool | Answer | How sure we are | Break-even hours a week | Return, low | Return, base | Return, high |
|---|---|---|---|---|---|---|
| Chat assistant A | Consolidate | Low | 0.118 to 0.236 (base 0.164) | -$877 | $7,768 | $30,630 |
| Chat assistant B | Consolidate | Low | 0.103 to 0.207 (base 0.143) | -$415 | $4,824 | $18,680 |
| Coding assistant | Keep | High | 0.154 to 0.308 (base 0.213) | $809 | $4,706 | $13,669 |
| Image generator | Right-size | Low | 0.11 to 0.22 (base 0.152) | -$164 | $685 | $3,792 |
| Meeting transcriber | Investigate | High | 0.073 to 0.145 (base 0.1) | -$202 | $2,587 | $12,795 |
| Support-desk AI add-on | Right-size | High | 0.185 to 0.371 (base 0.257) | $227 | $4,416 | $16,887 |
Break-even hours are the hours a week each active user must save for the tool to pay for itself. They run from the dearest hour to the cheapest. Return is the value of the time saved, less the cost of the seats. It is a range, because the hours saved are my guesses. An hour of time costs $45, $65 or $90 (low, base, high), and a month is 4.33 weeks.
Why, in one line each
- Chat assistant A: Consolidate. The same people use two tools for the same work.
- Chat assistant B: Consolidate. The same people use two tools for the same work.
- Coding assistant: Keep. None of the other four answers applies. It pays for itself at my base guesses and people keep what it makes. It overlaps no other tool, and fewer than one seat in ten is idle.
- Image generator: Right-size. Many seats are idle, never used, or belong to people who left.
- Meeting transcriber: Investigate. People rarely keep what the tool makes, so the time saved cannot be trusted.
- Support-desk AI add-on: Right-size. Many seats are idle, never used, or belong to people who left.
Money that is not earning its keep
- Idle seats (certain): $480 to $520 a month, from 13 to 14 of 116 seats.
- Consolidation savings (a choice): $0 to $315 a month, up to 9 seats. It is never added to the idle figure.
- All 116 paid seats cost $4,285 a month.
What failed
My first checks found every problem I planted, so I added harder cases, and the checks missed some. The story is in "What failed" on the page, in the case study and in the technical log.
My judgement
What I would check first. I would compare these figures with actual running costs before anyone acts, starting with unused seats. Then I would check whether team sizes or tool prices could change in the coming months.
Where I am least sure. I am least sure of the hours-saved guesses, the tools marked Low, and any team whose work a log cannot show. A team's workload can also surge or vanish with the market, which no usage log predicts.
What would change my mind. Output kept far more often, or a team that cannot work without a tool, would move a verdict. So would a different hourly cost, or running costs that rise or fall sharply.
The meeting transcriber. I would ask its teams why they need it, because people rarely keep what it makes. If the need is real, I would test a free option, or one built and owned in-house.
The people side. I would ask the team before any seat goes, and show everyone the numbers. People should help shape the decision, not have it done to them.
Page 1 of 1