How to cost an automation before you build it
To cost an automation you need four numbers: how often the process runs, how long it takes, what that time costs, and what building plus maintaining it costs. The first three you can measure in an afternoon by asking whoever does the work. The fourth is where most business cases fall apart, because maintenance is almost always left out.
How do you measure the current state?
Ask the person who actually runs the process to keep a simple log for a week: date, how long, what went wrong. A week is enough to see the spread, and the spread matters more than the average. A process that normally takes five minutes but forty when something deviates has its value in exactly those forty.
Which time should be counted?
It is not only handling time. Count the interruption: being pulled out of something else, switching context, coming back. That cost is real even though it is uncomfortable to put a number on. Count waiting time too, meaning how long the case sits idle, if it affects the customer.
Which costs get forgotten?
Build cost is the easy part. These are the ones usually missing from the case:
- Maintenance. A flow needs review whenever a connected system updates. Budget a few hours a month.
- Model calls. Billed per token and scaling with volume. Small but not zero, and easy to lose control of without a ceiling.
- The review step. If ten percent goes to manual checking, you have removed ninety percent of the time, not all of it.
- The learning period. Accuracy is lower in the first weeks and somebody has to watch every run.
- Access and permissions. Often what stretches the timeline, and it is somebody else’s time.
What does the whole calculation look like?
Four figures in, four out. Produce them in this order and it shows early if it does not carry.
| Line | How to calculate it | Common mistake |
|---|---|---|
| Frequency | Runs per month, from the system | Estimated rather than measured |
| Time per run | A week of logging, by whoever does it | The manager’s guess is used |
| Hourly cost | Full employment cost, not salary | Only salary counted |
| Interruption | The context switch around the task | Left out entirely |
| Build | Quoted after mapping | Quoted before mapping |
| Maintenance | A few hours a month | Set to zero |
| Model calls | Text per run times volume | Assumed negligible |
| Review | Share going to manual checking | Assumed to be zero |
What payback period is reasonable?
If the automation does not pay back within twelve months, I do not build it. That is a deliberately hard threshold. The reason is that a process taking three years to pay back usually changes before then, and you have paid for a flow that no longer fits. Above two years the answer is almost always to simplify the process first instead.
Is there a reason to build when the maths is weak?
Sometimes the argument is not time but risk: manual entry producing errors a customer finds, or the process existing only in one person’s head. Those are legitimate reasons to automate even when the hourly maths is weak. But be honest that this is the argument you are running on, or the project gets measured against a saving it was never meant to deliver.
Common questions
What is a reasonable payback period for an AI automation?
I do not build automations that fail to pay back within twelve months. That is a deliberately hard threshold: a process taking three years to pay back usually changes before then, and you have paid for a flow that no longer fits. Be sceptical of vendors promising a specific percentage return before they have seen your process, because that figure cannot have been calculated.
What do the model calls cost to run?
Less than most people expect, but not zero, and they scale with volume. The cost is driven by how much text is sent per run rather than by the number of runs. A flow that sends whole documents every time gets expensive; one that sends only the retrieved passages does not. Set a daily ceiling from the start, so a runaway flow is caught by an alert rather than by the invoice.
What does maintaining a flow cost?
Budget a few hours a month per flow. It goes on connected systems updating, formats changing, and somebody reviewing the queue. It is the line most often missing from business cases and the one that most often makes them fail to hold.
More in the same area
AI automation 101: what it is and when it pays off
What AI automation actually is, where the line sits against ordinary automation, and how to decide whether a process is worth building.
Seven processes that are almost always worth automating
Concrete workflows where AI automation tends to pay off, and what makes those particular ones suitable.
