AI Automation: How to Build an Honest Business Case
Most automation calculations are too optimistic. What saved time is really worth, and the three line items that are almost always missing.
"That saves us 20 hours a month." The sentence comes up in almost every automation conversation, and it is rarely wrong. It is still useless as a business case — because it leaves the decisive question open: what happens to those 20 hours?
The thinking error: saved time is not saved money
If an automation frees up 20 hours a month, cash only appears if somebody is subsequently paid less. In small and mid-sized teams that is almost never the case. The person is still employed, the salary is unchanged — costs do not go down.
What does appear is capacity: time that used to go into copy-paste and is now available for something else. The value of that capacity is real, but it is a different line item from a cost reduction. That is why our AI automation ROI calculator deliberately reports a capacity value — freed time priced at the internal hourly rate — rather than calling it "savings".
The distinction is not pedantry. It determines who you can show the calculation to. Promising a management team "we save €24,000 a year" when the same salaries are paid at year end costs credibility — on the next project.
The three line items that are almost always missing
1. Build time. A workflow that runs reliably is rarely built in two hours. Realistically it involves design, build, a test run on real data, and one correction loop. Those hours belong in the calculation, not in a footnote.
2. Ongoing maintenance. Automations break when interfaces change, a field gets renamed, or a vendor adjusts its API behaviour. Without a maintenance line item you are budgeting for a system that does not exist.
3. The remaining manual work. Almost no process is 100% automated. Realistic is 70 to 90%; the rest stays manual — including the edge cases that used to be handled in passing and now stand out as exceptions.
Include those three and the result looks more sober. It also survives the follow-up question six months later, which is the whole point.
What makes a good candidate
Not every process is worth automating. The combination that almost always pays off:
- high frequency — daily or several times a week, not once a quarter
- clear rules — if the answer is "it depends", the process is not ready yet
- stable data sources — a form or an API, not an unstructured email
- noticeable pain — processes nobody minds will not be missed once automated either
The classic mistake is starting with the most complex process because that is where the biggest potential sits. Starting with the most boring one makes more sense: high frequency, clear rules, low risk. It produces the experience that makes the second automation faster and safer.
Frequently asked questions
At what level of savings is automation worth it?
There is no universal threshold, but a workable rule of thumb: if the build does not pay for itself through the capacity value within six to twelve months, the process is either too infrequent or too complex.
Do I need AI, or is classic automation enough?
Very often classic automation is enough. AI models earn their keep where unstructured text has to be interpreted — classifying free text, summarising content, routing enquiries. "If form, then CRM record" needs no model.
What about data protection?
For any automation that hands personal data to an external service, the processing agreement and the data flow need to be settled up front — before the build, not after. That belongs in the business case, because it takes time.
How do I measure afterwards whether it worked?
Record the baseline before you start: cycle time, error rate, number of cases per week. Without a before value nothing can be demonstrated afterwards — and the business case stays an assertion.
What you can do today
- Write down the three processes your team complains about most.
- Note for each: how often per week? How long per run? Who does it?
- Run the capacity value through the ROI calculator — with an honest automation rate, not 100%.
- Subtract build and maintenance effort. What remains is the defensible part.
- Start with the most boring of the three, not the biggest.
If a candidate survives that you would rather not build yourself: how we set up workflows like these is described under AI & automation — or talk it through in a first call.