The short answer
Compare benefits you can actually use with the total cost over the same period. Time saved becomes a cash saving only when an expense is genuinely removed.
Measure the complete process
Record the volume, turnaround time and cost of the current process. The AI trial must include corrections and cases handled manually. Testing only easy cases overstates the benefit.
- Benefits: redeployable time, avoided expenditure, demonstrated improvements in quality or contribution margin.
- Initial costs: data preparation, development, integration and training.
- Recurring costs: models, infrastructure, supervision, maintenance and checks.
A fictional example over twelve months
Assume 200 hours released each month, after corrections, valued at €30/hour. If 50% of that time can be redeployed, useful capacity represents 200 × 30 × 50% × 12 = €36,000 a year.
With an initial investment of €20,000 and annual operating costs of €10,000, the first-year cost is €30,000. The estimated capacity-based ROI is therefore: (36,000 − 30,000) / 30,000 × 100 = 20%.
This scenario assumes constant activity and twelve months of use. It describes neither a client result nor €36,000 of cash savings. With only 25% of time redeployed, the same calculation gives −40%.
Test several assumptions
Build a conservative and a central scenario. Include ramp-up, adoption and overlooked costs. Separate measured benefits from assumptions. Do not count the same hours twice as both savings and additional output.
Then set a success threshold and a review date. An AI assessment helps decide what to measure before committing to development.
For your decision
Measure the current process and specify how the time saved will be used.
