Before you buy an AI tool, find out where the hours go

AI has gone mainstream in German business. In Bitkom's September 2026 survey, 57% of companies with 20+ employees now use AI — up from 36% a year earlier. Yet the same survey shows the problems that follow: among companies already using AI, 66% name data-protection requirements as a hurdle and 54% say costs are hard to calculate.

That second number is the one I care about. Costs are hard to calculate when nobody measured the problem first.

Why "tool first" fails

The usual sequence: someone sees a demo, the company buys licences, three people try it, and six months later nobody can say what it saved. Not because AI doesn't work — because nobody defined what "working" means.

The automation work I led at VMware saved around 72,000 engineering hours a year. It worked because it targeted real operational pain — repetitive work, avoidable mistakes — not because of any particular tool.

The four-column method

Take one week and fill in a simple table for every recurring task your team does:

Task	Minutes each	Times per month	What a mistake costs
Copy order data from email into ERP	6	400	Wrong delivery, angry customer
Prepare monthly sales report	240	1	Wrong decisions
Set up a new employee's accounts	90	3	Day one without access

Multiply minutes by frequency. The first row is 40 hours a month — a quarter of a full-time job — spent on copy and paste.

Then sort, don't buy

For each high-cost task, ask three questions:

1. Is it rule-based? If the same input always gives the same output, you probably don't need AI at all. Classic automation is cheaper, faster and easier to audit.
2. Does it need judgment on messy input? Reading free-text emails, sorting documents, drafting replies — this is where AI earns its place.
3. Who checks the result? Every AI step needs a defined human checkpoint for what matters. That's also how you stay on the right side of the EU AI Act.

Sometimes the honest answer is a better spreadsheet, not AI.

What you end up with

A ranked list: the five tasks that cost you the most hours, whether each needs AI or plain automation, and a rough cost to fix. That list is worth more than any licence — because it tells you which licence, if any, to buy.

You can do this yourself with the table above. If you'd rather have someone who has done it at scale run it in two weeks, that's exactly what my automation audit is.

---

Sources

1. Bitkom — Erstmals nutzt die Mehrheit der Unternehmen KI (14 Sept 2026)
2. IW Köln — Ziele, Nutzen-Kosten-Verhältnis und Hemmnisse Künstlicher Intelligenz in der deutschen Wirtschaft