How to Find the One Task AI Should Do First (Before You Buy Anything)
Not every task is worth handing to AI. The best first one is boring, repeatable, and easy to check. Here's how to spot it without buying a thing.
Most teams don't have an AI problem. They have a "where do we start" problem.
(There are too many options, so they pick none.)
It's the same reason people freeze in a big store. Two kinds of toothpaste, you decide in five seconds. Five thousand kinds, you walk out with none. AI right now is the five-thousand-toothpaste aisle. The job isn't to be more ambitious. It's to narrow the choice down to the one task worth starting with.
Here's the good news. The best first task is almost always the most boring one. After sixteen years at winship labs, I've never once started a team on the flashy use case. We start on the thing nobody brags about, because that's the thing that actually sticks.
Start with what's eating the most time, not what sounds impressive
A principle I lean on: don't guess where AI helps, watch where the week actually goes. For one week, have your team note the tasks they do over and over. You're not looking for the hard, creative work. You're looking for the repetitive middle: the research pulls, the first-draft emails, the routing, the formatting nobody enjoys.
The flashy demo is the trap. The boring repeat is the opportunity.
Here's the honest catch, though. Watching where the week goes feels like overhead nobody has time for, and the task you land on won't impress anyone in the room. The harder part isn't finding it. It's resisting the shiny use case everybody wants to demo and starting with the dull one instead.
And there's plenty to find. McKinsey estimates about 30% of the activities in 60% of all occupations are automatable, mostly the repetitive middle of a job, not the whole job (McKinsey).
The Boring-Task Test
Not everything qualifies, and that's the point. I call it the Boring-Task Test, and a task passes when you can answer yes to three things:
→ Does it happen often? Daily or weekly beats once a quarter.
→ Does it follow the same steps every time? If you can write the steps down, AI can learn them.
→ Can a person quickly tell if the output is right? You want a clear check, not a black box.
Picture this. A sales coordinator researches every new account by hand before the first call: company news in Google, recent funding on Crunchbase, who the buyer reports to on LinkedIn. Call it twenty minutes a pop, a dozen times a week, the same steps every time, and anyone can glance at the result and tell if it's right. That's the one. It hits all three.
If a task fails those three, leave it alone for now. Forcing AI onto judgment-heavy or one-off work is exactly how projects stall and people decide AI "doesn't work for us."
The audit ranks your list in five minutes
You can do this by hand, or you can shortcut it. The free AI Audit runs 22 questions in about five minutes and scores where you're actually ready, across technology, process, and people, on the same 30/70 split we run every engagement on.
From there you pick one task, build it as a small AI workflow with a human checking the output, and prove it before you touch anything else. If you'd rather have someone rank the list and build the first one with you, that's what a fractional head of AI does, instead of handing you a roadmap and leaving.
No rush on any of it. Start with the boring task you already do every week. The unglamorous one is the one that lasts.
