AI Costing Prompts · By task
Quantify the cost of unused capacity
Give an AI model your capacity and cost figures and get the rate per hour at practical capacity, the cost of the capacity you actually used, and the cost of the capacity that sat idle. The prompt also explains why spreading idle cost across products quietly destroys your product costing.
In short
Provide the machine count, theoretical hours, a practical capacity factor, total monthly cost and actual productive hours. The model computes theoretical and practical capacity, sets the cost rate per hour at practical capacity, then splits total cost into used and unused. The unused figure is the cost of idle capacity, and it should be reported as a period cost, not buried in product cost where it triggers a death spiral.
What the prompt is doing
This is the capacity logic that sits underneath time-driven activity-based costing. The crucial move is to set the cost rate using practical capacity, the hours you can realistically run after maintenance, setups and normal downtime, rather than theoretical capacity or last month's actual hours. Practical capacity is usually around 80 to 85% of theoretical, and using it gives a stable rate that does not lurch every time volume changes. See capacity costing for the full method.
The reason this matters is the death spiral. If you divide total cost by only the hours you happened to use, the rate rises whenever volume falls. That higher rate inflates product cost, which makes products look unprofitable, which leads to dropping them, which lowers volume further, which raises the rate again. The fix is to cost products at the practical-capacity rate and report the cost of unused capacity separately, as a visible management number rather than a hidden tax on the products that remain. This is the same discipline you build into a TDABC model.
The prompt
You are a cost accountant. Work only from the data I give you. Do not invent any numbers. Flag any assumptions clearly. Production cell: Machines: 4 Available hours per machine per month (theoretical): 360 Practical capacity factor: 85% Total monthly cost: 48,000 EUR Actual productive hours last month: 980 Do the following: 1. Calculate theoretical capacity in hours. 2. Calculate practical capacity in hours. 3. Calculate the capacity cost rate per hour (at practical capacity). 4. Calculate the cost of used capacity. 5. Calculate the cost of unused capacity and unused capacity as a % of practical capacity. 6. Explain why spreading the unused cost across products distorts product cost. 7. List any assumptions. Output a summary table.
A worked example
| Measure | Value |
|---|---|
| Theoretical capacity | 4 x 360 = 1,440 h |
| Practical capacity | 1,440 x 0.85 = 1,224 h |
| Capacity cost rate | 48,000 / 1,224 = 39.22 EUR/h |
| Used capacity cost | 980 x 39.22 = 38,431 EUR |
| Unused capacity cost | 48,000 - 38,431 = 9,569 EUR |
| Unused capacity | 244 h = 19.9% of practical |
| Utilisation | 80.1% of practical |
Product cost should be built at the practical-capacity rate of 39.22 EUR/h, with the 9,569 EUR of unused capacity reported as a separate period item.
If instead you spread the full 48,000 over only the 980 productive hours, the rate jumps to 48.98 EUR/h, a 24.9% inflation. That is the death spiral in numbers: higher unit cost makes a product look unprofitable, you drop it, volume falls, and the rate rises again on what is left. Keeping the rate anchored to practical capacity breaks the loop and turns idle cost into a visible management decision rather than a hidden distortion.
Assumption flagged: total cost is not split into fixed and variable here. If part of the 48,000 is genuinely variable, it falls when hours fall, so the true cost of idle capacity is lower. Treat the 9,569 EUR as an upper bound until the fixed and variable split is known.
What it costs you to run
The input runs to about 280 tokens. A full run with the calculations, the summary table and the death-spiral explanation costs a fraction of a cent on any current model. Re-running it for several cells or for different utilisation levels stays trivially cheap.
Token figures are approximate and vary by model and tokeniser.
The guardrail that matters
Work only from the data I give you. Do not invent any numbers. Flag any assumptions clearly.
Capacity prompts go wrong when a model assumes a practical capacity factor or a fixed-versus-variable split that you did not give it. This line keeps it inside your numbers and makes it state the upper-bound caveat instead of presenting one figure as certain. See how to stop AI inventing numbers.
When you need the real model
One cell is straightforward. A real plant has many cost pools, shared resources and time equations that route cost to the work that consumes it. Building that properly, so capacity is costed consistently across the whole operation, is the work behind a defensible TDABC model.