How to audit your first automation candidate in an afternoon

This is a working exercise, not a theory piece. Block out two hours and run it against one real workflow — not the one that sounds most impressive to automate, the one your team actually does every week.

The exercise

1. Write the workflow down, step by step, exactly as a human does it today. Not the ideal version — the actual one, including the workarounds. If step 4 is technically "check the CRM" but actually means "check the CRM, then double-check against the spreadsheet someone keeps because the CRM is sometimes wrong," write both parts down. The workarounds are usually where the real complexity lives, and they're the first thing an idealized process diagram hides.

2. Mark every step that requires a judgment call. Circle each place where the person doing it has to decide something rather than just execute. "Enter the invoice number" isn't a judgment call. "Decide whether this invoice looks suspicious enough to flag" is. Be honest here — it's tempting to undercount judgment calls because they make the workflow look less automatable, but undercounting them is exactly what produces automation projects that fail six months in.

3. For each judgment call, write the rule they're actually using. If you can't write the rule in one sentence, that step isn't ready to automate — it needs to be simplified or standardized first. This step does double duty: it either produces a rule clear enough to hand to a machine, or it reveals that the "rule" was never actually consistent, which is a process problem no amount of automation will fix.

4. Count how many times this workflow runs per week, and how long each run takes. Multiply. That's your real baseline — not an estimate, an actual number you can defend if someone asks where it came from.

5. Ask: what would "good enough" automation look like for the parts that ARE rule-based? Not "fully automated end-to-end" — just the mechanical middle, with a human still making the judgment calls at the start and end. This reframing is what separates a project that ships in weeks from one that stalls trying to handle every edge case before launch.

What a good result actually looks like

If you finish this and have a clear, rule-based middle section with a known weekly time cost, you have a real automation candidate — something you can scope, price, and actually build with confidence about the payoff.

If every step still needs a judgment call, you've found a training or process problem instead of an automation opportunity. That's a useful result too, just a different one. It means the workflow needs to be made consistent by the humans running it before a machine can run any part of it — and knowing that before you spend a dollar building anything is exactly what this exercise is for.

Run it against three or four candidate workflows and you'll usually find one clear winner — the one where step 3 was easy to answer and step 4's number was bigger than expected. Start there.