An invoice for manual work never arrives. Salaries arrive, rent arrives, the subscription arrives — but nobody bills you for the half hour a day someone spends copying orders from an inbox into a spreadsheet. That is why those tasks survive for years: nobody ever costed them.
Costing them takes five minutes, and the result is usually unpleasant. Below: the method, the four mistakes that turn the calculation into wishful thinking, and what to do when the answer comes back “it does not pay”.
You can run the numbers here: manual task cost calculator. It will also tell you to leave the task with a human — more on that at the end.
The arithmetic is not the hard part
Minutes per run × runs per day × working days per month × people ÷ 60 = hours per month. Multiply by the hourly cost for a month, by twelve for a year.
Twelve minutes, eight times a day, one person, twenty-one working days: 33.6 hours a month, 403 hours a year. At $30 an hour that is about $12,000 a year for something that feels like “ten minutes, tops”.
That feeling is exactly why nobody counts. Ten minutes is nothing. Four hundred hours is two and a half months of somebody’s working life.
An hour costs more than salary divided by hours
First mistake. A person costs more than their contract says.
Take what actually leaves the account for them each month: salary, employer taxes, the desk, the laptop, the share of paid leave. Divide by hours actually worked — not the nominal 160, but what is left after holiday and sick days, roughly 145–150.
That usually lands about half again above “salary ÷ 160”. Count on salary alone and every number after it is understated by the same margin.
Automation does not remove the whole task
Second mistake, and the most common. People assume the task goes to zero after launch. It almost never does.
What stays is everything the rules did not cover: an order with a typo in the address, a customer who writes “same as last time”, a supplier who changed their export format. And supervision stays: somebody glances once a day to check it all went through. Across our own work, the share genuinely taken off a person is usually seventy to ninety per cent. A hundred happens only where the task has no exceptions at all — and those tasks were usually automated long ago.
That is why the calculator has a “taken off the person” field, and why it starts at eighty. If someone promises a hundred, ask what happens to an order whose address is misspelled.
Owning it costs money too
Third mistake. Automation is not a purchase, it is ownership.
Other people’s systems change without warning: a marketplace redesigns a page, a bank changes its statement format, a supplier switches protocol. Anything that reads somebody else’s data needs attention a few times a year. Anything that lives entirely inside your own perimeter may need nothing for years.
The difference is fundamental and worth establishing before the work starts. Put it to your contractor as: how many outside systems does the result depend on, and what happens when each of them changes. If there is no answer, the cost of ownership is unknown and it is too early to talk about payback.
Freed hours are not money yet
Fourth mistake, and it is not an arithmetic one.
If, after automation, the person stays on the payroll doing what they did before minus this task, you have not saved money. You have freed time. It becomes money only when the time is used: taking more orders, not hiring the second person, ending the overtime.
This is common and reasonable — for a small business the bottleneck is usually people, not cash. Then automation buys throughput rather than savings: the same people handling twice the volume. But it has to be counted honestly and called by its real name, or in six months someone will look for a saving that was never there.
When it does not pay
“It does not pay” is a legitimate answer and it comes up more often than people expect. Three typical cases.
The task is rare. Five minutes once a week is four hours a year. No amount of arithmetic makes development worth it.
The task is cheap and full of exceptions. If automation takes off only half and upkeep eats a third of the gain, the payback period runs past the horizon on which the process will still look the same.
The process is about to change. Automating something that will be redesigned in six months means paying twice. The right move there is not automation but settling the process first.
There is a fourth case the calculator cannot see: the task can sometimes be removed rather than automated. A reconciliation often exists only because two departments keep the same list twice. Deleting the second list is cheaper than automating the comparison between them. We propose this regularly, and it is the one kind of work we have an interest in talking you out of.
What to do with the answer
Under a year — run it again with a higher budget and a lower share taken off. If it is still under a year, waiting costs more than doing it.
Two to three years — the question is no longer money but stability. Will the process still look like this? If yes, go ahead. If no, fix the process first.
It does not pay — then do not automate. That is an honest answer, and it is free.
Run your own numbers: manual task cost calculator. The numbers live in the page address, so the result can be shared as a link.
If it does pay back, tell us about the task — we will say what comes off the person and what has to stay.