Three questions decide whether something should be automated, and answering them takes about a minute.

The third question matters most and gets asked least. A task that runs weekly with messy inputs, where a mistake surfaces immediately, is a fine candidate. A task that runs weekly with clean inputs, where a mistake shows up in a client’s inbox, is not.

Answer those three and most work sorts itself into three piles.

Automate it fully: moving data between systems, pulling the same numbers into the same report every week, filing things, chasing overdue invoices. Anything where the correct output can be checked by a rule.

Automate it with a person on the end: first drafts, summaries, categorisation, anything where the output is judged rather than verified. The automation kills the blank page. A person still decides.

Leave it alone: anything where being wrong is expensive and quiet. Pricing. Contracts. Anything a client reads while assuming a person wrote it.

The trap in practice is the second pile pretending to be the first. Everybody reviews the output at the start. A month in, it has been right so often that reviewing feels like ceremony, so the checks quietly stop. The mistakes that follow are the ones nobody catches, because the process that would have caught them was retired by attrition rather than by decision.

Build the check into the workflow rather than into somebody’s discipline and that does not happen. A draft that lands in a review queue gets reviewed. A draft that lands in a send folder gets sent.

Work out what is worth automating ›››

All posts ›››