The monthly report is done by Tuesday midday. Three systems, two spreadsheets, one person who knows the order it all has to happen in. That is how it has worked for years, and it works well.
When someone asks at this point whether AI could not take it over, the answer is usually a yes with a shrug. The question that should come next is one hardly anyone asks: what does this report cost today?
Calculating what a workflow costs sounds like homework for the finance team. It is the one part of the exercise you cannot catch up on later. Once the workflow runs differently, the old number is gone, and with it any way of showing whether anything changed.
Why you calculate workflow costs beforehand
In our experience, AI plans in an SME start with a tool and end in an argument nobody can win. It feels faster. Proving it is another matter. Six months on, the management team sits in front of the question of whether the thing should stay, and the only solid number on the table is the licence cost.
That is not a measurement problem, it is a question of sequence. A number about the state beforehand can only be taken beforehand. After that, the people involved remember different versions of how long it used to take. The person who does the work remembers the worst month. The person who pays for it remembers the fastest one. Both mean it honestly, and neither memory is good enough to decide on.
Our own blog goes out in two languages without anyone being awake in the morning. It used to mean writing, translating, generating the image, uploading, checking, every step by hand and in that order. What we never wrote down is what the old way cost. We can say what runs today, but not by how much it got better. The gap this piece is about is one we left open ourselves. That is why we insist on it with clients, and why we keep being reminded of it at home.
The five items that make up the price of a workflow
We ask about the same five things in every first conversation. Not because the number has to be right to the franc, but because adding them up regularly makes something visible that nobody had seen that way before.
Working time
The obvious item, and the one that in our experience gets set too low. An hour of work costs you more than the wage: the employer contribution to AHV alone is 4.35 per cent of the relevant wages (Art. 13 AHVG), the Swiss old-age and survivors' insurance. On top of that come the contributions to disability and income compensation insurance and to unemployment insurance, half of each carried by the employer, the premium for occupational accidents and at least half of the occupational pension contributions.
Then a second correction. A full-time position does not contain 52 working weeks. The law requires at least four weeks of holiday per year of service, five for employees up to the age of 20 (Art. 329a CO). Add public holidays, sickness and training. Anyone who builds an internal hourly rate from annual salary divided by calendar hours is costing every workflow cheaper than it is.
Handovers
Every move from one person to the next costs twice: once handing over, once picking up. The person who passes on the report on Tuesday writes an email with it. The person who takes it over reads in, finds something unclear, asks back and waits.
Count the handovers in a workflow once, from the trigger to the result. In the workflows we have looked at, the number was usually higher than expected, and it explains more about the duration than the actual work does. Every handover needs a moment when the next person happens to have time. That is what the duration hangs on, not the speed of the work itself.
External costs
Everything that goes to third parties per run. The service provider who prepares the data, the agency for the same mailing every time, the translation service, the subscription that exists only for this one step.
This item is the easiest of them all, because it sits in the books. It still gets forgotten, because it lives in a different account from the salaries and therefore rarely shows up next to the working time.
Errors and rework
What happens when the workflow goes wrong, and how often does it go wrong? The wrong figure in the report, the invoice corrected after the fact, the quote that goes out with the wrong line item and comes back.
Errors cost twice: the repair, and the caution afterwards. In our experience, after a visible mistake somebody double-checks every run for months. That checking time belongs to the workflow, even though no description of it says so.
Waiting time
The time in which nothing happens and it still costs something. The order waiting for an approval. The quote that sits for three days because the person who has to sign it is away.
Waiting time is the item that time tracking structurally cannot see. What gets recorded is what someone is working on, not what is waiting for someone to work on it. In workflows that touch customers, it can be the most expensive of the five, because it hangs on revenue rather than on cost.
What the number does to the decision
You find the first three items in systems. You find the last two only by asking the people who do the work. Those are the two that tip the decision.
A workflow that looks harmless by working time can carry a substantial part of its annual value in rework and waiting time. And the other way round: a workflow that everyone complains about but that runs once a quarter is not a case for a rebuild, however unpleasant it feels. Frequency times cost per run decides, not annoyance per run.
This calculation belongs on a single sheet that the management team confirms. For us that sheet sits at the start of our first value project, four weeks in which one workflow is rebuilt and measured against exactly this number. Without the sheet, whether it worked ends up being a matter of interpretation. Ask us how that works if you want the detail.
The estimate that holds up in the meeting
Three of the five items you will estimate. That is fine, as long as you write it down that way.
An estimate holds when it comes from the person who does the work, when it is marked as an estimate, and when it gives a range instead of false precision. Two to four hours per run is a usable figure. Three hours and twelve minutes is a claim that falls at the first objection and takes the whole sheet with it.
What does not hold: a number taken from an industry benchmark. Calculating workflow costs means using your own figures. Outside averages sound more precise and are worthless for your decision.
Where this number stops
The number does not tell you whether the workflow is the right one. A number about recurring work says what that work costs, not whether it hangs on the business. The question before it is a different one, and we have written about it elsewhere: the best AI use case is in your income statement. Direction first, then the number.
The calculation is not a promise either. Saved hours do not turn into money by themselves. They turn into money when work disappears, when an external contract is no longer needed, or when the freed-up time lands where the business is stuck. Otherwise the saved quarter of an hour spreads across the day and ends up in the inbox.
And it does not replace the work that comes after the rebuild. A workflow that runs without people still needs somebody keeping an eye on it. We once took apart this invisible work behind AI agents. That work belongs in the same calculation, just on the other side.
How to spot a suitable workflow
Two things you can observe without measuring anything.
First: the workflow runs the same way every week or every month, and you can say what should come out at the end. Anything that feels different each time is judgement work and stays with people.
Second: the data for it sits in two or three systems, not in twelve, and it sits there digitally. A workflow that hangs on a binder full of paper is a filing project first and a case for AI second.
If both apply, it is worth calculating what this workflow costs before anyone rebuilds anything. You have the figures in the house, with the people who do the work today. Nobody has to invent anything here, which is why we do this calculation before an offer rather than after one.
The sheet that comes before anyone talks price
Three things belong on it, and nothing more is needed to start:
- The workflow, from which trigger to which result, and how often it runs per year.
- The five items with a figure next to each, estimates marked as estimates.
- The annual value, and next to it an honest sentence about which share of it realistically disappears.
It takes one session with the person who does the work today. It is the least spectacular part of the whole undertaking and the only one you cannot repeat later.
Which workflow came to mind first as you read this?
If you want to check whether it suits this calculation, write to us for a short first conversation. It costs nothing and ends with an assessment, not a quote.
Frequently asked questions
How do I calculate what a workflow costs?
Add up five items per run: working time at the full employer rate, handovers between people, external costs to third parties, errors and rework, and waiting time. Multiplied by how often it runs per year, that gives the annual value. You will estimate three of the five, which is fine as long as each estimate is marked as one.
Why does the measurement have to happen before the start?
Because the earlier state no longer exists afterwards. Once the workflow runs differently, the people involved remember different versions of how long it used to take. Without the number written down beforehand, whether the rebuild achieved anything ends up being a matter of interpretation.
Which workflow is a good one to start with?
One that repeats the same way every week or month with a result you can name, and whose data sits digitally in two or three systems. Workflows that feel different every time are judgement work and stay with people. If everything hangs on paper, it is a filing project first.
