IDSPIRATION STUDIO · GUIDE · ADOPTION
Before you roll out AI.
How to choose a first AI pilot your team will actually use, by starting with people and the work, not the tool.
THE SHORT ANSWER
Don’t start with a tool. Start with a task that is frequent, frustrating and forgiving of small mistakes. Involve the people who do that task, help them understand how the tool works before teaching them what to click, and run a small pilot with a clear human check. Then measure whether behavior changed, not whether people logged in.
01
Start with the task, not the tool.
Many AI rollouts begin with a purchase: licenses for everyone, a launch email, maybe a training session. Then everyone is left to work out what it’s for. Usage spikes, then fades.
A better starting question is: what part of the work do people wish took less of their time? The answer gives you a real problem to solve, and a group of people who already want it solved.
02
Use three filters.
Good first pilots pass all three:
| Filter | The question | Why it matters |
|---|---|---|
| Frequent | Does it happen every week, or every day? | Habits only form around things people do often |
| Frustrating | Do people complain about it? | Relief is the strongest reason to change |
| Forgiving | Is a small mistake easy to catch and fix? | People need room to learn without fear |
Drafting first versions, summarizing notes and sorting incoming requests often pass. Anything customer-facing, legal or financial usually fails the third filter, so save it for later.
03
Involve the people who do the work.
The people who do a task every day know where the time goes and what “good” looks like. Ask two or three of them to help design the pilot, including at least one skeptic. Skeptics find the real problems early, and when they come around, others follow.
These early volunteers often become your best advocates. A colleague showing how they actually use a tool is more convincing than any launch email.
04
Teach how it works before what to click.
Feature training goes out of date quickly, and it doesn’t help people decide when to use a tool. A basic understanding does: what AI is good at, where it makes things up, and why the same question can give different answers.
That understanding is what lets people judge for themselves, which is the difference between using a tool and adopting it.
FROM EXPERIENCE: UNDERSTANDING CREATES ADVOCATES
In a previous role where the approach to AI was to hand out tools and hope for the best, Steven ran AI literacy workshops from the basics up. Afterward, people from different teams started coming forward with their own ideas, which grew into an AI advocate program: colleagues building real use cases and sharing them with their own teams.
05
Run a small pilot with a human check.
Keep the first pilot small: one task, a handful of people, two to four weeks. Before it starts, agree on three things:
- What the tool does, and what it doesn’t.
- What a person always checks before anything is used or sent.
- What “better” means: time saved, fewer errors, less frustration.
A clear human check isn’t a sign of distrust. It’s what makes people comfortable trying something new.
06
Measure behavior, then decide.
Logins tell you who opened the tool. They don’t tell you whether the work changed. At the end of the pilot, talk to the people involved and look at four signals: use, habit, trust and value. Our adoption audit shows how.
Then make one of three decisions: scale it to more people, change it and run again, or stop it. Stopping a pilot that didn’t work is a good outcome. It’s far cheaper than rolling out a tool nobody uses.
About this guide: written by Steven Klimek from experience leading digital and customer change in large organizations. The three filters are a practical starting point, not a validated research instrument.