The calculation almost everyone gets wrong
The typical mistake is multiplying time saved by hourly cost and calling the result savings. That calculation ignores the human review every automation needs, the exceptions that stay manual and the costs of the system itself.
The right question has two parts: how much time capacity is freed and what freeing that capacity costs. Only by comparing both figures can you decide with judgement.
A hypothetical example, step by step
Imagine a process of 300 tasks a month, today taking 6 minutes each: 300 × 6 minutes = 1,800 minutes, that is, 30 gross hours a month. This is a hypothetical figure explaining the method, not a customer outcome.
With automation, each task arrives classified and with a draft, and human review takes 2 minutes: 300 × 2 minutes = 600 minutes, that is, 10 hours. The gross difference is 20 hours a month, before deducting exceptions, errors and the time spent supervising the system.
If 10% of cases (30 cases) are exceptions taking 10 additional minutes each, beyond the review already counted, subtract 5 more hours (30 × 10 minutes). Freed capacity in this hypothetical example is then 15 monthly hours, not 30, before deducting errors and system supervision time. This is the breakdown to request in any proposal.
Freed time is not saved money
Fifteen freed hours a month are capacity, not cash. They only become money where those hours go to value-creating work: serving more customers, selling more or avoiding a hire that would otherwise be needed. Actual cash cost reduction, potential extra revenue and the resulting margin are different figures: estimate them separately.
If freed time dissolves into the working day with no destination, automation improves comfort but not the bottom line. Decide beforehand where those hours will go and who will verify it.
Repeat the calculation with your data
For a comparable period, record four inputs: number of tasks, current minutes per task, review minutes and exceptional cases with their additional time. Add build, running, maintenance and supervision costs separately.
You can express freed capacity as current hours minus review hours and exception hours, then subtract the relevant supervision time. Keep freed hours, euro cost and the destination of that time separate; that is what makes the decision reviewable with observed data.
- Current hours = tasks × current minutes ÷ 60
- Review hours = tasks × review minutes ÷ 60
- Exception hours = exceptional cases × extra minutes ÷ 60
- System and supervision costs, separate from freed time
The costs that do exist
Every automation has four costs: building it (initial project), running it (AI models, third-party services), maintaining it (changes, errors, updates) and supervising it (each month's human review).
Ask every proposal for this breakdown with written assumptions: task volume, exception share and review minutes. Without written assumptions you can neither compare proposals nor challenge deviations.
- Build: an initial project with agreed scope
- Running: AI models and third-party services by volume
- Maintenance: changes, errors and updates
- Supervision: human review and exception handling
How to estimate your case
Measure your current process for two weeks: number of tasks, minutes per task and the share of unusual cases. With those three numbers, the example's method gives you an honest gross estimate.
Then subtract review and exceptions in hours, and compare the system's euro costs separately against the value of those hours at their destination. If the balance stays positive and you know where the freed hours will go, you have a possible case; uncertainty remains until the system is observed in use. If not, you still have something valuable: the certainty of not investing where it does not pay.