To manage a team that uses AI, hold one fifteen-minute conversation every week and ask four things: what the team used AI for, what it got wrong, what people are unsure they are allowed to do, and what they should stop doing by hand. Answer the first question yourself, first. Then act on one answer before the next meeting.
Across the business teams I train, the pattern is the same. The people who decide whether AI sticks are not the owner or the IT supplier. They are the team leads, the office managers and the heads of department, because they are the ones standing there when somebody asks, "Am I allowed to do this?" This article gives them a routine for answering, a sheet to copy, and a table that says which answers they can settle alone.
Why does the manager matter more than the policy?
A policy says what is allowed. A manager decides what happens on Monday when it does not cover the question. The evidence points the same way on both sides of the Atlantic, and it also shows how little help managers are getting.
Start with the US. Gallup's quarterly survey of working adults (43,262 responses from US adults aged 18 and over who work full or part time, fielded February 4 to 19 and May 6 to 20, 2026, published July 21, 2026) found that 48% of employees whose manager actively supports AI use were engaged, against 30% of those whose manager does not. Where employees used AI at least weekly, their organization had a clear plan, and their manager supported them, engagement reached 53%. Gallup measured engagement, not output, and a survey like this cannot show that manager support caused the gap. Gallup's own summary of the manager's role is blunt: "Gallup has consistently found that managers are the primary way employees experience their organization."
Now the managers' side. The Chartered Management Institute surveyed 1,019 UK managers in February 2026 and published the results in June 2026. Just 12% rated themselves "very confident" in their ability to support their teams' AI adoption. Only 27% said their organization provides comprehensive AI usage policies. And only 13% strongly agreed that their senior leaders actively experiment with the tools themselves.
Put the two together and you see the gap. The person with the most influence over whether a team uses AI well is the person least likely to feel ready, and least likely to see the people above them going first. Neither survey is built around small firms, so read them as a pattern, not as a measure of your business. That reading is mine, not the surveys'.
What is the Friday Four?
The Friday Four is a fifteen-minute team conversation, once a week, in which a manager asks four things: what did you use AI for, what did it get wrong, what are you unsure you are allowed to do, and what should we stop doing by hand? The manager answers the first question first. Each conversation ends with one action, one owner and one date.
I offer it as my suggested way of thinking, not as research. It works because each question finds something a policy cannot.
| Question | What it finds | What the manager does with the answer |
|---|---|---|
| 1. What did you use AI for this week? (manager answers first) | Real use, which is usually narrower or wider than you assumed | Write each job down. Keep the ones that worked. |
| 2. What did it get wrong, or nearly get wrong? | Errors caught before a customer saw them | Fix the check, not the person (see the Loop Test below) |
| 3. What are you unsure you are allowed to do? | Gaps in permission | Answer from the written rules today, or take the question to the owner by Monday |
| 4. What should we stop doing by hand? | The next job to hand over | Choose one job and one person to try it |
Four rules keep the conversation honest:
- Fifteen minutes, same time every week. A routine that slips to "when we have time" disappears in a month.
- The manager goes first on question one. If you want people to admit what they tried, show them it is safe. Do it with a small, real job from your own week, even one that went badly.
- No blame on question two. An error reported is a check that worked. An error hidden is the one that reaches a customer.
- One action, one owner, one date. A conversation without an action is a meeting. Write it down before anyone leaves.
Here is the sheet to copy. One row per week:
| Week of | Who used AI, and for what | What went wrong or nearly did | What people are unsure about | What to stop doing by hand | The one action | Owner | Done by |
|---|---|---|---|---|---|---|---|
Keep it in a shared folder where the whole team can read it. By week four it is the start of your team's own rules, written in their words.
What does a Friday Four sound like in practice?
Take a hypothetical 18-person bookkeeping and payroll practice. Dana leads a team of five. In week one, she goes first and says she used AI to draft three payment reminders, then checked each against the client list before sending. Two colleagues used it to summarize call notes. Sam used it to write a spreadsheet formula.
On question two, Sam admits the formula worked on the test rows and failed on a client with a blank cell. Nobody lost money, because Sam checked three clients by hand. On question three, Priya asks whether she can paste a client's payroll file into the tool to summarize it. Dana does not know. She says so, and says she will have an answer by Monday.
On question four, the team picks monthly chasing emails. The one action is Dana's: take the payroll file question to the owner. Until the owner answers, the team's rule is that no client files go into any AI tool. That rule is now written on the sheet.
Four weeks later, the sheet holds four rows, two rules and one job handed over. Nobody wrote a policy. The team has the front half of one, and a manager who can show it to the owner as evidence, not opinion. If the owner wants to turn those rows into a firm-wide document, Does a small business need an AI policy? and the UK small business AI policy template show how.
What I See When I Train Business Teams
Across the business teams I train, the questions that stall adoption are rarely technical. They are "Am I allowed?" and "What do I do when it is wrong?" Both arrive at the manager's desk. The manager has no answer and often no one to ask, so the reply becomes "just be careful." That sentence does quiet damage. Cautious people stop using AI, and confident people carry on without telling anyone.
The weekly conversation changes that. It turns a private worry into a decision the team can see, and it gives the manager something to say besides "be careful." As I put it on my business page, "The hard part of AI at work isn't the technology. It's helping people use it well, safely and every day. That's what an educator does." The business page sets out how that works in practice.
What should a manager decide alone, and what goes to the owner?
A manager can settle anything that stays inside the team's own work and touches no customer, employee or new tool. Everything else goes up. Use this table to sort the answers from question three before Friday ends.
| Decide it in the room | Take it to the owner this week | Stop the work today |
|---|---|---|
| Which routine job the team tries next | A new tool, app or subscription | A mistake that has already reached a customer |
| Who checks each result before it goes out | Customer or employee data going into an AI tool | AI used to rank, score or reject people |
| The wording of a reply or note template | A rule that would apply to other teams | Anything the team was already told never to do |
| How to describe the job on the weekly sheet | What customers are told about AI use |
Three of those rows have fuller answers elsewhere on this site. For data going into tools, see shadow AI in a small business. For telling customers, see AI disclosure in a small business. For anything touching hiring, see using AI to screen job applicants.
Question two deserves one more test. When someone says, "I check everything it writes," ask what I call the Loop Test: "The Loop Test is three questions I ask before accepting 'a human checks it' as an answer: Would that person notice a wrong result without being told to look for it? Would they notice in time to stop it mattering? And if they missed it, could the work be put back? Oversight that survives all three is a safeguard. Oversight that fails any of them is a signature." If Sam's formula had run on a full client list without a hand check, it would have failed the second question.
What if you are not confident with AI yourself?
You do not need to be good at it. You need to go first and be honest. CMI's 12% figure says most managers are in your position, and a team can tell the difference between a manager who is learning and one who is pretending.
Run the Confidence Test on yourself before your first Friday. The Confidence Test is three questions I ask a member of staff, not their leader, to find out whether AI confidence exists in an organization: Do you know what you are allowed to use it for? Have you used it on your own work this week? If it got something wrong, would you tell someone? Confidence is when all three answers are yes from the people least likely to give them.
The Friday Four is the weekly version of those three questions. Question three is the first, question one is the second, and question two is the third. If your own answers are no, no, and not sure, you have your first question one: say so, then try one small job before the next meeting.
If nobody has anything to say in the first two weeks, do not push harder in the room. Ask each person privately which part of their week they would drop, and start there. The Chore Rule covers that conversation.
What should a manager do this week?
Do five small things in order. None needs a budget or a project.
- Monday: Find your business's written AI rules. If there are none, write three lines on what you think is allowed and send them to the owner for a yes or no. The reply becomes your first rule.
- Tuesday: Use AI yourself on one real job from your own week. Note what it got wrong. This is your answer to question one.
- Wednesday: Book fifteen minutes with your team, the same time for the next four weeks.
- Thursday: Copy the sheet above into a shared folder and fill in your own row first.
- Friday: Run the conversation. Close with one action, one owner and one date.
Then, on the Monday after, do the action before anything else. If the same answer to question three keeps coming back, it is probably a permission gap, not a skill gap, and the Gap Test in what AI training does your team need shows how to tell the two apart. After six weeks, check whether use is real and regular with the reach, rhythm and role check.
Where to go next
If you want your managers to run conversations like this with confidence, that is what my AI training for teams is built for. If you would rather see where your business stands first, the Business AI Readiness Scorecard takes about five minutes. And if you are planning a managers' away day or a business event, my business keynotes are written for that audience.
Sources and further reading
- Gallup, "Employee Engagement Remains Flat as AI Adoption Accelerates", July 21, 2026 (US adults aged 18 and over working full or part time; 43,262 responses; fieldwork February 4 to 19 and May 6 to 20, 2026): gallup.com
- Chartered Management Institute, "The Management Imperative in AI Adoption", June 2026 (survey of 1,019 UK managers, February 2026): managers.org.uk
- Dan Fitzpatrick, AI for business teams, The AI Educator
Dan Fitzpatrick, The AI Educator, is a Forbes contributor, international keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries. About Dan
