How to pick the first thing to automate (without wasting money)

Most automation projects fail for a boring reason: the wrong task went first.

People pick the task that annoys them most. That is rarely the task that pays back fastest. Annoyance and cost are different things — the job that makes you sigh twice a week is cheaper than the quiet one that eats four hours every single morning without anyone complaining.

Here is the test we use in the audit.

1. Does it happen on a schedule you can predict?

Automation is good at "every time X happens, do Y". It is bad at "sometimes, depending". If you cannot finish the sentence "every time ___ happens, we always ___", the task is not ready yet. That does not mean never — it usually means the process needs a decision written down first.

2. Does it produce something you can sell?

Sort your tasks into two piles: work a customer would pay for, and work that merely has to happen. Invoicing has to happen. Nobody pays extra for it. Automate the second pile first, always. It is lower risk and the savings are unambiguous.

3. How many people touch it?

A task that three people each spend an hour on is worth more than one person's three-hour task, because you also remove the handoffs — and handoffs are where things get dropped, duplicated, and re-keyed.

4. What happens if it silently stops?

This is the question people skip. Automations fail quietly. If nobody would notice for a week, you need monitoring before you need automation. We treat a workflow that suddenly processes nothing as a failure, not a quiet day.

What not to automate

Judgement work. Anything where the right answer depends on context a human holds and a rule cannot capture: pricing an unusual job, handling an upset customer, deciding whether a supplier's excuse is credible.

You can support those with automation — pull the information together, draft the reply — but the decision stays human. Teams that automate judgement end up quietly turning the system off, which is the most expensive outcome of all.

The short version

Start with the highest-volume, most predictable, least judgement-heavy task that produces nothing you could sell. It is usually unglamorous. It is usually data entry, chasing, or copying between two systems that should have been talking to each other years ago.