# AI payback sheet — editable worksheet

Workflow: [ ]  Owner: [ ]  Observation period: [ ]

Use measured inputs where possible. Mark guesses as estimates. Time recovered is capacity, not automatically cash saved.

## Baseline and pilot
| Input | Baseline | Pilot | Evidence |
| --- | --- | --- | --- |
| Tasks per week | | | |
| Minutes handling each task | | | |
| Minutes reviewing each task | | | |
| Minutes correcting each task | | | |
| Error rate | | | |
| Loaded hourly labour cost | | | |

## Calculation
- Baseline minutes per task = handling + reviewing + correcting baseline minutes.
- Pilot minutes per task = handling + reviewing + correcting pilot minutes.
- Net hours recovered weekly = tasks per week × (baseline minutes − pilot minutes) / 60.
- Annual capacity value = net hours recovered weekly × working weeks per year × loaded hourly cost.
- Annual running cost = licences + model usage + maintenance + monitoring + other recurring costs.
- Annual net value = annual capacity value − annual running cost.
- Payback months = upfront implementation cost / (annual net value / 12), only if annual net value is positive.

## Costs and assumptions
Upfront implementation: [ ]
Working weeks per year: [ ]
Licence cost/year: [ ]
Usage cost/year: [ ]
Maintenance and monitoring/year: [ ]
Other recurring costs/year: [ ]

## Sensitivity and decision
Recalculate with half the task volume, twice the review time and twice the usage cost.
Low / expected / high annual net value: [ ] / [ ] / [ ]
Can recovered capacity actually be used? [how]
Decision: [pilot / proceed / defer]  Owner: [ ]  Next review: [ ]