AI for SMBs: Start instead of wait

The perfect time for AI never comes. Waiting until all data is clean and all processes documented means starting in three years, usually never.

The pragmatic entry point: pick a process that is frequent, expensive and easy to describe. That's where the first real return appears.

More important than the model is the environment: access to the right data, clean interfaces, a person who checks results. This isn't a tech project, it's an organisational project.

Start small: a pilot with a clear success criterion, a runtime of a few weeks and a decision rule for what happens next.

What we never do: introduce AI just to have AI. What we do: automate where it pays off, and honestly say where it doesn't.