The process everyone complains about is rarely the one worth automating first. Irritation tracks how unpleasant work feels; savings track how much of it there is. A tedious task done four times a month is a bad first project no matter how much it is hated, and a dull one done four hundred times is usually a good one.
The arithmetic is deliberately crude at this stage: how many times a month does this happen, how many minutes does one instance take, and what does an hour of the person doing it cost fully loaded. Multiply. If the annual figure is not a multiple of a plausible build cost, the process is not your first project, and no amount of sophistication in the build will change that.
Volume. Under a few hundred instances a month, most automations will not repay a build inside a year.
Minutes per instance. Ten minutes of copying between systems is a better target than two hours of genuine thinking.
Loaded cost. Salary plus employment costs plus the tooling that person needs — not the headline salary.
Growth. A process whose volume is growing is worth more than the same process flat, because you are also buying the headcount you will not add.
A saving that is arithmetically real can still be unbuildable. The question that separates the two is whether somebody in the business can state the decision rules out loud. Not perfectly, and not in writing yet — but if you sit with the person who does the work and ask "how do you decide?", you should get an answer with structure in it rather than "you just know after a while".
"You just know" is not a reason to give up; it is a signal that the first phase of the project is documentation, and that the documentation has value on its own. Several of our audits have ended with a client automating nothing and still getting their money back in clarity, because writing the rules down exposed three exceptions nobody had agreed on.
A useful test: ask two people who do the same job to describe the same decision separately. If their answers differ materially, you have found the real first project, and it is not an automation.
Automations that finish by writing a record into a system you already run are far easier to prove than automations that finish by producing a document a human then interprets. The first kind produces a countable result; the second produces something that still needs judgement, and the saving becomes an argument.
This is why invoice intake, ticket triage and lead enrichment are such common first projects. Each of them ends in a system of record — the ERP, the helpdesk, the CRM — with a row you can count, an error rate you can measure and a before-and-after you can put in front of a finance director without a debate about methodology.
Take four or five candidate processes and score each on four axes. Do it on one page, with the person who owns each process in the room, and accept that the ranking will surprise somebody.
Annual manual cost. Volume × minutes × loaded rate. The size of the prize.
Rule clarity. Can the decisions be written down this week? If not, subtract heavily.
System access. Does every system in the chain have an API or a supported export? A missing one adds weeks.
Blast radius. What happens if it gets one wrong? A misrouted ticket is recoverable; a wrongly posted payment is not. Start where mistakes are cheap.
The highest-scoring process is your first project. The second-highest is usually the better second project than whatever the first one suggests, because the two will share integrations you have already built.
The most common failure we see is not choosing the wrong process — it is choosing the right process and automating half of it. Half a process moves work rather than removing it: the queue simply forms at a different desk, and the saving that was projected never appears in anyone actual workload.
End to end means from the trigger that starts the work to the record that ends it, including the exception paths. Exceptions are usually 10–20% of instances and 60–80% of the difficulty, which is exactly why an automation that skips them looks cheap in a proposal and expensive in production.
Turning the crude arithmetic above into a number that survives a finance review.
Put your own volume, minutes and rate in and see the payback period.
The free mini-audit, the paid audit, the pilot, and what each stage produces.