Most teams pick their first AI project the way they pick a conference talk: whatever sounds most impressive. That’s how you end up with a demo that wows the board and never ships.
Boring beats bold
The automations that pay back fastest share three traits. They happen often, they follow a recognisable pattern, and someone can tell you exactly how long they take today.
Frequency: at least 50 times a week
Pattern: a person could write the instructions on one page
Measurable: you know today’s time, cost or error rate
Score every candidate
We list every repetitive task the team mentions in a one-hour workshop, then score each from 1 to 5 on volume, rule clarity, data access and risk. Anything above 15 is a candidate for the first sprint.
If you can’t measure the before, you can’t celebrate the after.
Start with a human in the loop
Your first automation should propose, not decide. Let the agent draft, classify or prepare — and have a person approve. After two weeks of clean approvals, you’ll have the data to remove the training wheels.
Do this once, measure it honestly, and the second project will sell itself.
Written by
Maya Okafor
Previously led operations at a Series C logistics company. Has automated her own inbox three times.
