When a team gets an automation builder, the first thing they try to build is a churn intervention. Something clever, conditional, multi-branch. It takes three weeks and it gets switched off in month two, because nobody trusts what it is doing.
The workflow worth doing first is duller, and it works.
Find the thing someone does every week by hand
In most product teams there is a person who spends part of Monday morning assembling the same view. They pull last week’s numbers, compare them to the week before, note anything that moved past a threshold they hold in their head, and paste the result into a document or a channel.
That person has, without meaning to, written a specification. The thresholds are in their head, the sources are in their browser history, and the output format is in last week’s message. All of it is recoverable in a half hour conversation.
Why this one first
It is verifiable. Run the automation alongside the human for two weeks and compare. Any divergence is a bug you can see, not a hypothetical effect on a conversion rate you will argue about for a quarter.
It has an obvious owner. The person who was doing it by hand knows immediately when the output is wrong, and cares enough to say so.
It pays back visibly. Ninety minutes a week is not a rounding error to the person who was spending it, and it accrues from the first run.
Critically, it teaches the team the tool on a task where being wrong costs a slightly odd message in a channel, rather than an email to two hundred customers.
What usually surfaces
Two things, reliably.
The first is that the thresholds were never written down and are not consistent. The person applies judgement, and the judgement is usually good, but it is not a rule. Turning it into a rule is the real work, and it is worth doing whether or not you automate anything.
The second is that one of the numbers has been quietly wrong for months and nobody noticed, because a human reading it every week smoothed over the oddity. A machine will not. This is irritating, and it is the most valuable thing the exercise produces.
Then the second one
Once the weekly review runs itself, the next candidates announce themselves. They are the other things that happen on a cadence: the quarterly access review, the monthly invoice reconciliation, the alert someone forwards to the same three people every time.
The churn intervention can wait. It will be built better by a team that has already found out where its own thresholds are.
