Generative AI is being pitched for almost every internal workflow right now. Some of it is genuinely valuable; a lot of it is expensive theatre.
Good fits
High-volume, repetitive, judgement-light tasks with a clear source of truth — support ticket triage, document summarisation, first-draft content.
Poor fits
Low-volume, high-stakes decisions with no ground truth to evaluate against. Automating these usually creates more review work than it saves.
A simple test
If you can't describe how you'd measure whether the AI is "right," it's not ready to be automated — it's ready for a human-in-the-loop pilot instead.
