The question always comes in the same order. You start with the task that annoys you — keying in invoices, chasing unpaid bills, reconciling the bank — and you look for a tool to get rid of it. That is a poor starting point. The task that annoys you most is almost never the one that pays back best, and the gap between the two runs into thousands of euros.
The reflex that costs you
What annoys is not what costs. A tiresome task done once a week weighs ten times less than a mundane one done twenty times a day. Annoyance measures friction, not volume — and volume is what decides. The first sort is to set aside what you feel and start counting.
Four criteria, in this order
Volume first: how many times a day, multiplied by the duration. Then how alike the occurrences are: the more they resemble one another, the better the automation holds. Then where the data sits. And finally, what happens if the system gets it wrong. These four criteria, in this order, explain almost every decision.
Volume rules out more projects than you would think
A task taking ten minutes a day amounts to forty hours a year. At €25 an hour, that is €1,000. No custom build pays back on that: the cost of keeping it running eats the saving within months. Two hours a day, by contrast, come to 480 hours and around €12,000. The same technical work, a case twelve times stronger.
Where the data sits kills more projects than complexity
This is the criterion most people underestimate. If the information already sits in a business system, almost anything is possible. If it sits in emails or PDFs, it has to be read first, which adds a layer of technology and cost. And if it sits on paper, it has to be produced before anything can be automated — groundwork that is often shorter than it looks, but that comes first.
And what you never get round to doing
This is the category nobody looks at. Chasing unpaid invoices when a spare moment turns up. Monitoring tenders, never started. Reactivating old customers, put off for two years. These tasks cost no hours, because they do not happen — they cost revenue that never comes in. No payback calculation sees them, and they are often the ones that pay best.
Where to start in practice
Take three tasks, not one. For each, note how many times a day it is done, how long it takes, and where the data sits. You will almost always find that only one of the three stands out clearly — and that it is not the one you had in mind. What matters is the ranking, not the precision of the sums.
We have built a free diagnostic that does the sorting for you: four questions to price the annual cost of a task, six more to work out what is recoverable, with which technology and over how many months. You can run it on three tasks in a row and compare. And if it concludes that none of them is worth automating, it says that too.