How to get skeptical employees to adopt AI: one owner, one task, a visible result.

To get skeptical employees to adopt AI, do three things. Name one person who owns the rollout. Select one task that the skeptics dislike, run the AI on it for two weeks and show the before and after numbers to everybody. Then decide, and tell the team, what happens with the time that AI saves. An order by email does not change behavior. A visible result on their own work does.

A short video from 2026 described a patent attorney whose lawyers were “kicking and screaming” about AI until their productivity rose by a large amount. The advice in the video was simple. If you back down, the tools stay unused. If you give an order by email, you get compliance on paper. Instead, name one owner, measure one task that the skeptics dislike, and show the result.

This page expands that advice into a method.

Why are employees skeptical?

Skepticism usually has a reason. Find the reason before you try to change it.

Each reason needs a different answer. A training course does not fix fear for the job. A policy does not fix a bad first experience.

Step 1: Name one owner

One person owns the rollout for one team. Not a committee. Not “everyone.”

The owner:

• Selects the task.
• Measures the baseline.
• Collects the feedback each week.
• Fixes problems or escalates them.
• Shows the result to the team.

The best owner is often a respected person inside the team, not a manager from outside. If the skeptics respect the owner, they will listen to the result.

Step 2: Select a task that the skeptics dislike

Do not start with the task that AI does best in a demo. Start with a task that the team hates: repetitive, boring, slow. Examples:

• Writing the same type of summary after each meeting.
• Copying data from emails into a system.
• Searching old documents for a clause or a number.
• Drafting standard replies to common questions.
• Formatting reports.

When AI removes a task that people dislike, it feels like help, not like a threat.

Avoid tasks that the team sees as the core of their expertise. Experts protect those tasks. You can come to them later, when trust exists.

Step 3: Measure before

Measure the task before you start. Use real data where possible:

• How many times does the task occur each week?
• How long does it take each time?
• How many errors or corrections occur?

If no system data exists, ask three people to time the task for one week. Write the numbers down. Show them to the team. The team must agree that the numbers are fair.

Step 4: Run for two weeks

Run the AI on the task for two weeks. Use these rules:

• Use a tool that fits the task. Not a general chat window for everything. Configure the tool with the templates, documents and examples that the task needs.
• Put it inside the daily tool where possible. Each extra step reduces use.
• Keep the human in control. The AI drafts. The person checks and sends. People must stay responsible for their work.
• Collect every problem. Make it easy to report a bad output. Fix the common ones during the two weeks.
• Include at least one skeptic. If the skeptics only hear about the result, they will doubt it. If they see it on their own work, they will believe it.

Step 5: Show the result

After two weeks, measure the same numbers. Show them to the whole team. Show both the good and the bad:

Then ask the team: “What would make this better?” Use their answers for the next version.

Step 6: Decide what the saved time is for

This step is the one that most companies skip. It is the main reason that people fear AI.

If the saved time becomes more work with no change, people learn that AI is bad for them. They use it less, or they hide that they use it.

Decide one of these options, and tell the team clearly:

• The time goes to higher-value work that the team wants to do, for example more client contact or better quality.
• The team takes on growth without longer hours.
• People who automate their own tasks get credit in their reviews or promotions. Some companies now reward employees who find ways to automate their work.
• Overtime goes down.

Be honest. If the company plans to reduce the team, say so, and explain how. People can accept a hard truth more easily than a rumor.

How do I train people?

Training on general AI skills does little for adoption. Training on the real task does much more.

• Use real work. Train with the team’s own documents and cases, not with generic examples.
• Keep it short. One hour, then practice on real work.
• Show the limits. Show where the AI fails, and how to check its output. This builds trust with experts.
• Give time. Give people protected time in the first weeks to learn. Without time, busy people go back to the old way.
• Use peers. People learn more from a colleague who uses the tool well than from a trainer.

What makes resistance worse?

Avoid these mistakes:

• An order by email. “From Monday, everyone uses AI.” People comply on paper and change nothing.
• Usage targets. Counting prompts per person measures activity, not value. People will produce prompts to meet the target.
• A ban, then a sudden push. If the company first banned AI, then suddenly demands it, people do not know what to believe. Explain the change.
• A slow approval path. If the approved tool takes months to arrive, people use unapproved tools with their own accounts. One fractional leader put it simply: if your approval path is slower than your unapproved path, people use the phone in their pocket.
• Silence about jobs. If leaders do not talk about the effect on jobs, people assume the worst.
• Leaders who do not use it. If managers ask staff to use AI but do not use it themselves, the message is weak.

What rules do employees need?

People need clear and short rules. Unclear rules cause both fear and misuse. A one-page policy is usually sufficient:

• Which tools are approved.
• Which data can go into each tool, and which data cannot.
• Who checks the output before it goes to a customer or a regulator.
• How to ask for a new tool, and how fast the answer comes.
• Who to ask when unsure.

Publish it before the rollout. Explain it in the first training.

How do I handle the expert who refuses?

Some experts refuse to use AI. Often they have a good reason: they see errors that others miss.

Do not force them. Use their expertise:

• Ask them to build the eval set: the examples and the good answers that the AI must match.
• Ask them to review a sample of AI outputs each week.
• Ask them to write the rules for the cases where the AI must not act alone.

Experts who help to define quality often become the strongest supporters. They trust a system that they helped to test.

What should managers say?

Managers often make adoption harder with the wrong words. Use clear, honest statements.

Say this:

• “We will test this on one task for two weeks. We will show you the numbers.”
• “You stay responsible for your work. The AI drafts, you decide.”
• “The time that this saves will go to [a specific thing].”
• “If it does not help, we will stop it.”
• “Tell me when it gives a bad answer. That is useful.”

Do not say this:

• “AI will make us all more productive.” (Too general. People hear: more work.)
• “Everyone must use AI from now on.” (An order without a reason.)
• “This will not change any jobs.” (Unless you are sure. People remember broken promises.)
• “It is easy.” (When it is not easy for them, they feel stupid.)

How do I measure adoption that matters?

Do not measure only logins or prompts. Measure these instead:

• Task coverage. The share of the selected task that people now do with AI.
• Time per task. Before and after.
• Quality. Errors, corrections and complaints, before and after.
• Voluntary use. Do people use it for other tasks without being asked?
• Feedback volume. People who use a tool send suggestions. Silence often means no use.

A hypothetical example

A finance team of eight people rejected an AI tool. They had tried a general chat tool and it gave wrong numbers. The controller named a senior accountant as the owner.

The owner selected the task that the team hated most: the monthly commentary on budget variances. It took about two days of the team’s time each month. The owner configured a tool with the previous commentaries, the chart of accounts and a template. The AI drafted. The accountants checked every number and edited the text.

After one month-end close, the time for the commentary fell from about 16 hours to about 6. Two accountants found errors in the AI draft. They added rules that fixed them. The controller told the team that the saved time would go to a cash forecast that the CFO had wanted for a year. The next month, two accountants asked to use the tool for the audit preparation.

Next step

Name one owner and one disliked task for one team this week. Measure the task before you start. If the larger problem is that AI shows no result anywhere in the company, read why companies see no ROI from AI. If you need help with the first rollout and the rules, read how my fractional AI CTO engagement works.