Every company we walk into already knows which process everyone hates. It gets mentioned in the first ten minutes, usually with a laugh and a specific person's name attached. It is almost never the right thing to automate first.
The loudest process is loud because it's painful, and it's usually painful because it's ambiguous — lots of judgment, lots of exceptions, lots of undocumented rules. That combination makes it expensive to automate and easy to get wrong in public. Failing there poisons the well for everything after it.
Score candidates on four axes
We rank every candidate workflow on volume, time per instance, cost of an error, and rule clarity. The first three tell you how much value is on the table. The fourth tells you how hard it will be to capture.
A process running 4,000 times a month at six minutes each with well-defined rules beats a twice-weekly judgment call that everyone complains about, every time — even though nobody will thank you for picking it.
Optimize the first project for proof, not for value
The first automation's real job is to make the second one easier to fund. That means it needs to finish, work visibly, and produce a number someone will repeat in a meeting.
Pick something you can ship in six weeks with a measurable before-and-after. Boring is a feature here.
What we look for
- High repetition with low variance in the inputs
- A clear owner who can answer rule questions same-day
- An existing measurement you can compare against
- Failure that is annoying rather than catastrophic
- A structural connection to the next two candidates on the list
The infrastructure dividend
That last point matters more than it looks. The first project pays for the plumbing — the integration to the ERP, the document extraction pipeline, the observability, the deployment path. If you choose it well, the second project is 40% cheaper and the third is 60% cheaper, because most of what they need already exists.
Choose it badly and each project pays full freight again. We have seen companies conclude that automation 'doesn't work for us' after three unrelated pilots that shared nothing.