An AI rollout plan should start with one task your team already does each week. Pick that task. Agree what a good result looks like. Give people the inputs, the time and a person to check the work. After a set trial, decide to keep it, change it or stop.
My view is simple. If your team went back to the old way, that tells you something. Either the tool can't do the task, or nobody set up the work around it. Those two problems need different fixes, so find out which one you have first.
This plan is for an owner who has already paid for AI tools. It works for a team of five or fifty.
Why do teams go back to the old way?
Most teams go back because the tool arrived before the task did. People got a login. Nobody showed them which job it was for, or how to check the result.
Training is often missing too. A 2024 Microsoft survey asked 31,000 office workers in 31 markets, Malaysia among them. Of those who use AI at work, only 39% said their firm had trained them.
The same survey found 78% of AI users bring their own tools to work. So your team may already use AI, but not the tool you paid for. Ask them before you plan anything.
Is it the tool or the rollout?
Look at what happened before you decide. The signs point to one cause or the other.
Tool or rollout?
Those two problems need different fixes. Find out which one you have first.
| What you see | What it points to | What to do |
|---|---|---|
| The output is wrong even with good inputs | The tool can't do this task | Test another tool or drop the task |
| The output is fine, but nobody uses it | The rollout | Fix the time, owner or steps |
| People don't know which job it's for | The rollout | Name one task and one owner |
| People use their own AI tool instead | The tool you chose | Ask why theirs works better |
| The output needs data people can't share | The setup | Sort out what data is allowed |
If you can't tell, run the plan below on one task. A small trial will show you.
An AI rollout plan for one task
Use only the steps your task needs. A simple task may need three.
- Pick one task. Choose work that repeats each week and has a clear output. Drafting replies, summing up call notes and first drafts of quotes are common choices. Need help choosing? This guide to where AI goes first puts a number on each option.
- Write down a good result. Describe what the output must contain. Note what would make it wrong. The person who does the task today should help write this.
- Gather the inputs. List what the tool needs to do the job. That could be a price list, past emails or a product sheet. Check that you're allowed to use that data in the tool.
- Give it an owner and time. Name one person who runs the trial and collects problems. Give the team time to learn it, as well as a login.
- Keep a person checking. Someone reads each output before it goes out. They note what they fixed and why.
- Set an end date and decide. At the end, look at the notes. Keep it, change it or stop.
One task, one trial
- Pick one task
- Write down a good result
- Gather the inputs
- Give it an owner and time
- Keep a person checking
- Set an end date and decide
Three calls Keep it, change it or stop.
How it could look on one task
Here's how this could look. The company and details are invented to show the steps.
A small parts seller wants AI to draft email replies to price questions. Today, one sales coordinator answers them. She helps list what a good reply needs. That's the right price, stock level and delivery date.
The tool needs the current price list and stock report. The owner checks that these can go into the tool. The coordinator owns the trial and reads every draft before it's sent.
After two weeks, her notes show a pattern. The drafts get the tone right but miss stock levels. The stock report wasn't connected. That's a setup problem, not a people problem or a tool problem. The owner connects the report and runs two more weeks. Then they decide.
Made-up example: a parts seller
- Weeks 1–2Drafts miss stock levelsThe tone is right
- CauseStock report not connected
- Weeks 3–4Report connectedTwo more weeks
- ThenDecideKeep, change or stop
A setup problem, not a people problem or a tool problem. Connect the report, run two more weeks, then decide.
What should you measure?
Measure what the task needs, not how many people logged in. A login count tells you who opened the tool. It doesn't tell you whether the work got better.
Track three things during the trial:
- How many outputs were used, changed or thrown away
- What the checker had to fix, in their own words
- Whether the task now takes more or less of the team's time
Write these down each week. They're what you'll use to decide.
What if your team still won't use it?
Ask them why, and listen to the answer. People are rarely the problem. They may not trust the output, or the tool may add steps to their day.
Fix what they name, if you can. If the tool adds work and saves none, stop the trial. Stopping is a fine result. It costs less than paying for a tool nobody uses.
Thinking of buying a new tool instead? Check it first. These questions to ask an AI or software vendor show what's included before you sign.
Questions owners ask next
Do we need AI rules before we start?
You need a few clear rules before the trial starts. Say which data can go into the tool and who checks the output. A long policy can come later.
Should we train everyone at once?
No. Train the people in the trial on the one task. Wider training makes more sense once you know the task works.
How long should an AI trial run?
Run it long enough to see the task done many times. Set the end date before you start, so the trial doesn't drift.
What if my team worries AI will replace them?
Say plainly what the tool will and won't do. In this plan, a person checks the work and makes the call. The tool handles the repeat work in one task.
Who should own the rollout?
The person who does the task today, or their manager. They know what a good result looks like. An outside expert can help set it up, but can't judge the output for you.
Should we buy more tools if this one fails?
Not yet. Find out why it failed first. If the task was wrong, a new tool won't help. A clear process sometimes fixes the problem with no new tool. This guide to building a follow-up system shows that for sales follow-up.
Start with one task this week
Pick the task. Write down what a good result looks like. Name the person who'll check it, and set the end date.
My free newsletter covers AI for daily work. You'll see how to pick tools, check their work and decide what earns its place. I send it when I have a lesson worth sharing.
