Most decisions about automation are made on gut feeling. Somebody thinks invoice handling takes an unreasonable amount of time, somebody else thinks it is running fine, and then it is settled by whoever talks loudest in the meeting.
That is unnecessary, because the calculation is simple. It takes ten minutes and it produces a number you can argue about, which is the whole point.
The four numbers
You need four things and nothing more.
- How often. The number of times a week the task is done.
- How long. How long it takes each time, measured rather than guessed.
- Who. The monthly salary of the person who actually does it.
- How much goes away. What share can realistically be handled without a person.
Step one: turn the salary into an hourly cost
The monthly salary is not what the hour costs you. On top of it sit employer contributions and holiday pay, and across the year there are fewer working weeks than fifty-two.
We calculate it like this:
hourly cost = monthly salary × (1 + uplift) × 12 / (working weeks × 40)
We set the uplift at 42 per cent, which is the Swedish employer contribution of 31.42 per cent plus holiday pay. We set the working weeks at 46, that is 52 minus holiday and public holidays. Both can be questioned, and both should be written out so that anyone who wants to can change them.
An employee on SEK 36,000 a month then costs SEK 333 an hour. Not 225, which is what the salary divided by 160 would give you.
Step two: count the hours, not the feeling
Say the task takes six hours a week. Across 46 working weeks that is 276 hours a year.
Then judge how much can realistically be taken away. Almost never a hundred per cent, because somebody still has to look at the exceptions. Set it at 70 per cent and you have 193 hours.
193 hours times SEK 333 is SEK 64,411 a year. Roughly SEK 5,400 a month.
Now you have a number to decide on. If the build costs less than that and lasts more than a year, the answer is usually yes. That is the whole analysis.
The three places where the calculation lies
1. The time is always underestimated
When people estimate how long a task takes, they count the task itself. They do not count being interrupted, hunting down the file, asking a colleague and working their way back into what they were doing.
Measure for a week instead of asking. Our experience is that measured time is consistently higher than estimated time, and that the gap is widest for the tasks people like least.
2. Freed-up time is not cash flow
This is the most important sentence in the article. The number you arrived at is the value of time freed up. It is not money appearing in the account.
If the person is still employed and still working full time, your cost has not fallen by a single krona. What has happened is that 193 hours moved from something nobody wanted to do to something else. That is usually worth a great deal, but it is a different thing from a saving, and anybody who confuses the two loses credibility the second the finance director asks where it shows up in the accounts.
So say which one it is. Either the time goes to something that grows revenue, or you genuinely do not replace the next person who leaves. Both are valid. Not choosing is not.
3. The automation has a cost of its own
A built workflow is not free to own. Something breaks when a supplier changes an interface, somebody has to look at the exceptions, and somebody has to be able to answer why the workflow did what it did.
Count that in. A reasonable rule of thumb is to put a smaller share of the freed-up time back as maintenance. If the calculation is so thin that it cannot survive that deduction, the workflow was the wrong candidate to begin with.
When the answer is no
We turn work down on this calculation, and that is not a pose. Four cases where you should not build anything:
- The task is done rarely. Something that happens four times a year will never earn back the build, however annoying it is.
- The rule keeps changing. If the exceptions outnumber the main case you are automating chaos, and then you just get to the wrong answer faster.
- Nobody owns the workflow. If there is nobody at your company who can say what the right outcome is, there is nobody who can confirm that the workflow got it right.
- The underlying work should not be done at all. The most common and most profitable conclusion from a mapping session is that the report nobody reads should stop being produced, not be automated.
Do it yourselves
Take the task somebody at your company sighs about on Mondays. Measure it for a week. Run the formula above. Deduct a fifth for maintenance.
If the number comes out larger than a build costs, you have a case. If it comes out smaller, you have saved yourselves a project, and that is a result too.
The calculator on the home page runs exactly this calculation with sliders, and writes out every assumption so you can change them.