Every automation proposal you have ever read contains the same sentence: this will save the team ten hours a week. It is almost never checked afterwards, and it is usually wrong in both directions. Sometimes badly overstated, occasionally understated by a factor of three because nobody counted the error correction.
The calculation is not complicated. It is just rarely done honestly.
Start with the loaded cost, not the salary
The hour you are buying back does not cost the hourly salary. It costs salary plus employer taxes, plus benefits, plus the share of software licences, equipment, management overhead, and unproductive time that the hour carries. Across most of the markets we work in, that multiplier lands between 1.25 and 1.4 on base salary.
We use a loaded rate of 38 dollars an hour as a default for administrative and coordination work in North America and Western Europe. If your own figure is different, use yours. The point is to use a real one rather than the number that makes the case look best.
Count the hours the process actually takes
Ask the person doing the work, not their manager. The gap is consistently large, and it is not because anybody is lying. Managers estimate the happy path. The person doing it counts the reruns, the chasing, the double checking, and the fifteen minutes lost every time the system logs them out.
Then add the three categories that get left out:
- Context switching. An interrupt driven task that takes four minutes costs closer to fifteen once you count the re-entry into whatever it interrupted.
- Error correction. If two percent of records need fixing, and fixing one takes twenty minutes, that is a real line in the calculation.
- Escalation. The cases that get passed to someone senior cost that person's rate, not the original one.
Subtract what the automation itself costs to run
This is the line most proposals omit entirely. An automation is not free after it is built. It costs:
| Cost | Typical range | Notes |
|---|---|---|
| Platform subscription | $20 to $500 a month | Task or operation based pricing grows with volume, sometimes faster than you expect |
| Model inference | $0 to $400 a month | Only if an AI step is involved. Routing simple cases to a smaller model usually cuts this by more than half |
| Maintenance | 2 to 6 hours a month | Third party APIs change. Credentials expire. Somebody renames a field |
| Monitoring | Included or $0 to $50 | If you are not paying for this in some form, you do not have it |
A workflow that saves twenty hours a month at 38 dollars is returning 760 dollars a month. If it costs 180 dollars a month in platform and maintenance, the net is 580. That is the number that belongs in the business case.
Then divide by the build cost
Payback period is build cost divided by monthly net return. A 2,400 dollar workflow returning 580 a month pays back in just over four months. Anything under six months is straightforwardly worth doing. Six to twelve is worth doing if the process is stable. Past eighteen months, the process will probably change before the automation pays for itself, and you should say so.
If the payback is over eighteen months, the honest recommendation is usually to not build it.
The two adjustments that change the answer
First, volume trend. If the process volume is growing twenty percent a year, the automation gets more valuable every month and the payback calculation understates it. If the volume is shrinking, the reverse.
Second, the value of the hour returned. An hour taken off a fully occupied senior person who then does higher value work is worth far more than the loaded rate suggests. An hour taken off someone who then does nothing in particular is worth close to zero. This is the single biggest reason automation projects disappoint: the hours are saved and then absorbed rather than reinvested.
Before you approve any automation, write down what the freed hours will be spent on. If you cannot answer, the saving is theoretical.
What we do with this
We run this calculation before quoting, and we tell you the result even when it argues against the work. A few times a year we talk a client out of a build because the numbers do not support it. That costs us a project and buys a relationship, which has been a good trade every time.
Written from work we have actually delivered. If your situation looks like the one described here, the quote form takes about two minutes and there is no sales sequence attached to it.
Get a free quote