One great use case per person.

Most AI plans start at the top of the organization and work down. Someone draws up a roadmap, a few tools get licensed, and a training day goes on the calendar. Six weeks later the logins are mostly idle, and the people who were supposed to change how they work are still doing it the old way.

There is a smaller goal that tends to go further: every person on the team finds one task that AI makes clearly better for them. It can be modest and unlike anything a colleague is doing, as long as it belongs to them and works well enough that they would miss it.

Why the big plan stalls.

An organization-wide plan has to describe work in general terms, because it covers everyone. "Use AI to improve customer communication" is a reasonable line in a plan and a useless instruction on a Tuesday morning. The person at the front desk of a clinic, the estimator at a manufacturer and the grants coordinator at a nonprofit all read that sentence and see a different job, or no job at all.

General goals also leave the hardest part to each person: figuring out where the tool fits into a day that is already full. Most people try it once on something vague, get a vague answer, and decide it isn't for them. Nobody failed in that story. The plan just never got down to the size of an actual task.

What counts as a great use case.

A great use case is specific enough that you could describe it in one sentence with a verb and an object. "Draft the first version of the weekly board update from my notes." "Turn the inspection checklist into a plain-language summary for the property owner." "Sort the shared inbox each morning into what needs me today and what can wait."

It is also frequent. A task someone does once a quarter won't build a habit, however much time it saves. Something that comes up daily or weekly gives the person enough repetitions to get good at it, and to notice when the output is wrong.

And it has to be one the person cares about. The best candidates are usually tasks people put off, or do late in the day when their attention is gone. When AI takes the first pass at that kind of work, the change shows up in how the week feels, which is what makes someone keep going.

A use case only counts once the person would be annoyed to lose it.

Why one per person adds up.

Say each person on a team of twelve has one task that works. Those twelve wins rarely stay separate, because people talk about what's working for them, usually in passing. The office manager mentions how she handles vendor emails now, and the program director asks how she set it up. That kind of exchange spreads practical knowledge faster than any slide deck we have seen.

Each person also learns the general skill through a task they already understand. Someone who has spent a month getting good drafts of meeting follow-ups has learned how much context to give and when a draft should be thrown away. Those lessons carry into the second and third use case without anyone having to teach them again.

The organization gets something useful too: a real inventory of where AI is helping, written in the language of the work. That list is a far better input for deciding what to build or buy next than a vendor's feature sheet. It also tells you which tasks several people found separately, and those are the ones worth turning into a shared process.

How to find each person's one.

Start with the person's week, not with the tool. Ask them to walk through the last few working days and name the tasks that took longer than they should have, or that they dreaded. Writing, summarizing, sorting and reformatting tend to come up, because those are the parts of a job that AI handles well enough to be useful on the first try.

Then pick one together, try it during the conversation if you can, and agree on a small test. Two weeks is usually enough. The person uses AI on that task every time it comes up and keeps a rough note of where it helped and where it got things wrong.

Some first picks won't work. The tool may not have the information it needs, or the task may depend on judgment that is hard to hand off. That's a normal result of a first try. Pick another task and run the same test. What you are after is a use case that survives two weeks of ordinary use.

A one-to-one coaching script.

Any team lead can run this in about half an hour. Treat it as a working session with the laptop open, and keep it off the performance-review calendar.

Opening (5 minutes)

1. "I'm not checking how much you use AI. I want to find one thing it could make easier for you."
2. "Walk me through last week. What took longer than it should have?"
3. "Which of those do you put off, or leave for the end of the day?"

Choosing (10 minutes)

1. "Of those, which one comes up at least every week?"
2. "What would a good first draft of that look like, if someone handed it to you?"
3. "What information would they need to write it?" (This becomes the context the person gives the tool.)

Trying it (10 minutes)

1. Run the task once together, using real but non-sensitive material.
2. "What's usable here, and what would you change?"
3. "Would you use this next time, as it is or with changes?"

Closing (5 minutes)

1. "For the next two weeks, try it every time this task comes up. Keep a two-line note of what worked and what didn't."
2. Put the follow-up on both calendars before you leave the room.
3. At the follow-up, ask one question first: "Would you miss it if it went away?" If the answer is yes, write the use case down in one sentence and add it to the team's list. If it's no, go back to step 2 and pick the next task.