The first automation you ship matters more than it should. Not because of what it saves, but because it decides whether anyone in the building will agree to a second one. Pick badly and the technology is not what gets blamed. You are.
How the first one usually gets picked
Someone picks the most annoying thing. Reasonable instinct, wrong result, because the most annoying thing is usually annoying precisely because it is full of judgment calls and exceptions. It is annoying because it is hard.
Or someone picks the most impressive thing, because a pilot ought to demonstrate ambition. Impressive things touch many systems, need many approvals, and take a quarter. By the time it lands, whoever sponsored it has moved on.
Both produce the same outcome: a long build, a soft result, and a room full of people who have concluded this is not for them.
Four tests
1. It happens often
Daily is ideal. Weekly is workable. Anything monthly or rarer will not accumulate enough evidence to convince anyone before attention moves elsewhere. Frequency is what turns a small saving into an argument.
Twenty minutes a day is eighty hours a year. A four hour monthly report is forty eight. The small daily one is worth more and is almost always easier.
2. It is boring
You want a task where the correct output is not a matter of opinion. Reformatting, extracting, routing, summarising to a fixed shape, checking one list against another. If two competent people would produce the same result, it is a good candidate. If they would produce different results and both be defensible, leave it alone for now.
3. Being wrong is cheap and visible
This is the test people skip. Given that output is around 70% right, your first workflow must be one where an error is caught immediately and costs almost nothing.
Drafting a reply somebody reads before sending: excellent. A wrong draft costs ten seconds. Posting to the ledger unattended: bad, and the error surfaces at month end when it is expensive.
4. Somebody will notice it is gone
There has to be a specific person whose Tuesday is better. Not a department. A person, who can say out loud that they used to spend the first hour of every day on this and now they do not.
That sentence is what buys you the second project. Not a percentage in a deck.
What usually passes
- Inbound enquiry triage. Read the form, pull out the structured bits, score it, route it. Frequent, boring, wrong is obvious, and the person who used to read every form notices immediately.
- Turning notes into a record. Call notes or job notes into a consistent summary on the account. High frequency, low stakes, checked at a glance.
- The reconciliation nobody enjoys. Comparing two lists that should agree and flagging what does not. Not fixing. Flagging.
- First draft follow ups. Drafted, never sent automatically. A human sends.
What to leave for later
- Anything that touches money unsupervised. Not never. Not first.
- Anything a customer sees without review. Same reasoning, different currency.
- Anything requiring your judgment as the owner. If it is a call only you can make, automating it is automating the wrong layer.
- Anything resting on the data problem you have not fixed. If your customer records are duplicated, start there instead.
Set the bar before you start
Write down, in advance, what success looks like. One number and one sentence. “Triage drops from forty minutes a day to under ten, and Dana says it is better.”
If you cannot say in advance what would count as working, you will not be able to say afterwards whether it worked. That is how these things end up in limbo.
Then actually go back and ask. That step is the one everybody skips, and it is the difference between a system that improves and a pilot nobody ever formally killed.
Small, frequent, boring, cheap to get wrong, and owned by a person who will tell you the truth about it. That is the whole brief for number one.
