"This process is a waste of time" is easy to say and hard to put in a budget. A number is easier to discuss. The good news is that for one process you do not need a study. You need about an hour, the person who does the work, and three inputs.
This article is deliberately narrow. If you want to find out which of many processes to look at first, measuring the cost of a manual process in one week is the right start, and the true cost of manual processes explains the wider picture. Here you already have one process in mind and need its price.
The formula
The yearly cost of a manual process has two parts: the routine work and the rework caused by mistakes.
Routine work = time per run x runs per month x hourly cost x 12
Rework = runs per month x error rate x time to fix one error x hourly cost x 12
Add them together. That is the whole model. Everything else in this article is about getting honest inputs and using the result well.
Getting the inputs in one hour
Sit down with the person who actually does the work, not only with their manager. Fifteen minutes each for these four questions:
- How long does one run take? Ask them to do one while you watch, or to recall the last three. Take the typical value, not the best one. Include the small steps people forget: opening the file, finding the right customer, copying the result somewhere else.
- How many runs are there in a month? Check a calendar, a mailbox or a system export instead of asking for a guess. Memory usually underestimates volume.
- How often does something go wrong, and what does it cost to fix? Count the last ten runs and how many needed a correction. Ask how long the correction took, including the time of the other person who had to be asked.
- What does an hour of this person's time cost? Use a loaded hourly cost: salary plus social and health contributions, equipment and a share of space and management. Finance can usually give you one number for the role. If you cannot get it, take the gross salary and add about a third, and write down that this is an estimate.
Write each input in a table with a column called "measured or estimated". That single column makes the final number far more credible.
A worked example (illustrative, not a client case)
The numbers below are invented to show the arithmetic. Yours will differ.
Imagine a company that enters supplier orders from emails into its ordering system by hand.
| Input | Value | Source |
|---|---|---|
| Time per order | 12 minutes | measured on three orders |
| Orders per month | 300 | system export |
| Error rate | 4 percent | counted over ten runs, estimated |
| Time to fix one error | 45 minutes | estimated |
| Loaded hourly cost | 32 | given by finance |
Routine work: 12 minutes x 300 orders is 3,600 minutes, which is 60 hours a month. At 32 per hour that is 1,920 a month.
Rework: 300 orders x 4 percent is 12 errors. At 45 minutes each that is 9 hours, or 288 a month.
Total: 2,208 a month, or 26,496 a year.
Now test how fragile it is. If the error rate is really 2 percent and a fix takes 30 minutes, the rework drops to 3 hours, or 96 a month, and the yearly total falls to about 24,200. If the volume is 250 orders instead of 300, the routine part falls by a sixth. The honest answer is therefore "roughly 22,000 to 27,000 a year, depending on how the estimated inputs turn out". A range like this is easier to defend than a single figure with false precision.
What the number does not include
The calculation above is conservative on purpose. It leaves out several things that are real but harder to prove:
- delays to the customer or to the next team while the work waits,
- the cost of a serious mistake, such as a wrong order that is shipped,
- the effect on the person, such as boredom, turnover and training time for a new colleague.
Mention these in the request as unpriced risks, but do not add guessed values to the total. A number that finance can reproduce from your table beats a bigger number they have to take on trust.
Using the number in a budget request
A cost on its own does not get a budget. The request needs four more things:
- The cost of the change. Include the one-time work and the running cost afterwards, not only the build. What a custom system costs in years two to five shows what is easy to forget, and for the contract side see fixed price or time and materials.
- What remains. Automation rarely removes a process completely. In the example, perhaps one order in ten still needs a person to check it. Subtract that remaining time from the saving.
- The payback time. Divide the one-time cost by the monthly saving. If the change costs 12,000 and saves about 1,700 a month after the remaining checks, it pays back in a little over seven months. That is a sentence a manager can use.
- The cheapest alternative. Often a clearer template, a required field or a small script removes most of the errors for a fraction of the price. If you show that you considered this, the request looks more trustworthy, not weaker. When the cheap fix is not enough, moving from a spreadsheet to a small internal tool in 30 days is a practical next level.
Finish the request with one decision, for example "approve a four-week pilot with a limit of 12,000", and say how you will measure the result: the same three inputs, measured again a month after the change.
What to do this week
Pick the one process people complain about most. Book an hour with the person who does it, fill in the table above and mark every input as measured or estimated. Calculate the yearly cost and the range. You will either have a number to bring to your finance team, or the discovery that the process costs less than it feels, which is also useful to know.