Summer is traditionally the season for strategy talks at our company. This year something stands out: hardly any mid-sized company still asks whether AI automation is coming. Everyone asks where to start. The answer is less spectacular than the hype suggests.

Don't start with the most impressive thing — start with the most annoying

The best starting point is rarely the process that shines on the board slide. It is the one everyone groans about: sorting and answering the inbox, assembling quotes from legacy data, checking and matching invoices, compiling reports from three systems. These tasks have three properties that make them ideal: they are frequent, they follow patterns, and nobody is emotionally attached to them. An automation success at a point like this creates more tailwind than any presentation.

Starting small doesn't mean thinking small

A proven approach in three steps: first, pick a single process and measure it honestly — how many cases, how much time, how many errors. Second, run a pilot for four to six weeks in which the AI prepares and a human approves; this builds trust and a clean quality comparison. Third, only then expand — to more volume, more autonomy or the next process. What we strongly advise against: starting with a company-wide “AI initiative” that plans for twelve months before it has completed a single case.

The data excuse no longer holds

“Our data isn't ready” is something we hear often — and mostly it is only half true. Modern AI systems cope surprisingly well with unstructured emails, PDFs and grown filing systems. Perfect data is not a prerequisite for getting started; it is the result of the first automation projects: where processes are automated, data discipline emerges almost by itself.

If you are looking for the one process where getting started pays off for you: this is exactly the search we do all the time — and we're happy to do it with you.