Yes, sometimes, but only if you build the handover. Sending one person on an AI course to teach everyone else works when that person has hours set aside to teach, colleagues practice on their own real tasks within a week, and someone checks a month later that the skill reached the last person. Without those three things, the knowledge usually stays with the one who went.
One seat is easier to schedule than twelve, which is why this is such a common plan. The trouble is that the course teaches one person to use AI, while the business needs twelve people to use it. The handover between those two numbers is where the plan lives or dies. Below are what four sources say about that handover, a three-question test to run before you book the seat, and a comparison of the three ways a small team usually learns.
Why do owners send one person and hope it spreads?
Because it is a reasonable instinct. The person who goes is usually the keen one, and the rest of the team is busy.
The UK government found the same pattern. Its evaluation of the Flexible AI Upskilling Fund, published on 28 January 2026 by the Department for Science, Innovation and Technology (DSIT) with Ipsos as evaluator, covered a grant scheme for UK professional and business services firms with 1 to 249 employees. It reports that it was common for businesses to train a small number of team members "with the expectation that they would deliver feedback sessions" to everyone else. That expectation is the plan, and in most firms it is the only plan.
What does the evidence say about teaching the team afterward?
I found no study that measures whether this works for AI in small firms, but four sources show where it tends to break: the teacher's time, the learners' practice, and checking the result.
The UK fund evaluation measured the plan, not the result. Its process and baseline phase ran from November 2024 to June 2025, so it records what firms expected rather than what changed after training. When its surveys ran in January and February 2025, 37% of the 73 successful applicants who answered had completed their training and 51% had it booked. A survey of about 105 of their employees found 82% said they can make use of AI-based tools, but only 46% said they can quality assure the outputs, and 18% did not use AI at all in their role. Among the 73 employers, 63% named limited AI skills, expertise or knowledge as a barrier. The evaluators note the samples are small and self-selected, so treat these as a snapshot of motivated firms in the UK, not a national measure. The 46% matters most here, because checking output is exactly the skill that is hardest to pass on in a feedback session.
A cascade in another field shows what dilution looks like. In a qualitative evaluation of a national suicide prevention training rollout in Scotland (Gask, Coupe and Green, BMC Health Services Research, 2019), 97 facilitators were trained in 2008 to 2011 to train others. They went on to train 1,155 participants in 148 group sessions, but only 25 of the 97 facilitators delivered any training during the project. The researchers interviewed 19 facilitators, 30 participants and 11 managers. The subject is different from AI at work, and a health service is not a 20-person firm, so I use it only for the mechanism: facilitators who were engaged and "confident in their ability to train" did far better than those who were not. Being trained is not the same as being ready to train.
Current UK guidance on AI upskilling points the other way. The employer guide from the Department for Work and Pensions and Skills England, part of "Skills for AI: What works for AI upskilling in the UK" (published 10 June 2026, with the HTML guide added on 27 July 2026), asks that "learners are given paid time or protected time to take part". It says hands-on learning should link "directly to real tasks", suggests "revisiting training after 3 and 6 months", and warns that training is less effective when leaders do not "model its use". It mentions cascade models only as one item in a ranking from its survey and says nothing about training one person to teach the rest. Its one nod to peer help is that coaching from local champions can help learners who lack digital experience.
US small businesses report the same squeeze on time. The Federal Reserve Bank of San Francisco's qualitative brief on AI adoption among small businesses (1 September 2026, drawing on the 2024 Small Business Credit Survey of US firms) reports firms facing "time and capacity constraints related to staff training". The page does not state the sample size. The hours were already short before anyone was asked to teach.
My reading, separate from the sources: a cascade is not wrong, it is under-built. The sources agree on what a training plan needs (protected time, practice on real work, follow-up), and a cascade that leaves those out is a hope, not a plan.
What is the Relay Test?
The Relay Test is the three-question check I suggest an owner runs before sending one person on an AI course to teach the rest. A relay is won or lost at the handovers, not by the fastest runner.
The Relay Test is three questions an owner asks before sending one person on an AI course to teach the rest: Has the person been given hours to teach, written into their week? Will each colleague use what they learn on a real task of their own within seven days? And in a month, will anyone check that the last person in the chain can do it without help from the first? Three yeses make a relay. Anything less is a hope that knowledge will travel.
Has the person been given hours to teach, written into their week? Knowing how to do something and being able to teach it are different skills, as anyone who has taught learns early. If teaching is added to a full job as a favor, it will be the first thing to go. Put the hours in the calendar before you book the seat.
Will each colleague use it on a real task of their own within seven days? Practice inside the first week is the difference between a session people enjoyed and a habit they keep. It is the third rung of the Four Rungs in AI training for small business teams: what actually works. Name each person's task in advance.
In a month, will anyone check that the last person in the chain can do it without help from the first? Ask the people at the end of the chain, not the person who went. Use the Confidence Test: "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 checking skill in particular needs its own drill; see how to train staff to catch AI mistakes at work.
Score it plainly. Three yeses, go ahead. Two, fix the missing one before you book. One or none, do not send one person alone; pick another approach from the table below.
Which way of training a small team fits your business?
There are three common routes, and each fits a different kind of team. This comparison of approaches was checked against the sources above on 4 November 2026. It compares ways of working, never products or fees.
| Approach | What the sources say | Fits when | Breaks when |
|---|---|---|---|
| Send one person, then cascade | Common among firms in the UK fund, with no outcome data yet. In the Scottish rollout, about a quarter of trained facilitators went on to train anyone | The team is small, jobs are similar, and the teacher has written-in hours | Teaching is squeezed into spare time, or the teacher is not the person others turn to |
| Train the whole team together on shared jobs | The DWP and Skills England guide asks for protected time, real tasks and follow-up at 3 and 6 months | Up to roughly 25 people do similar work and can be in the room together | The session is a one-off with no practice week and no follow-up |
| Short modules for everyone, with a named coach | The same guide favors short, stackable modules of 30 to 90 minutes and mentions coaching from local champions | People work shifts or in different roles and cannot meet at once | The coach has no time, or the modules are not tied to anyone's actual job |
The team sizes in the third column are my rule of thumb, not figures from the sources.
My verdict: if you send one person, treat the course as the first leg of a relay and pass the Relay Test first. If you cannot answer yes to all three questions, train the whole team on one shared job, or use short modules with a named coach, and use the cascade only for a later top-up. If you are not sure what your team needs taught, start with the Gap Test in what AI training does your team need.
What I See When I Train Business Teams
Across the business teams I train, the person who went on the course is usually the one who needed it least: already curious, already using AI. They come back full of ideas and find no hour to share them. The colleagues who most need to learn have the least time to be taught. This is a composite pattern, not a single story.
The fix is rarely a better course. It is a calendar entry, a named task for each colleague, and a date when the owner asks someone who did not go to show a result. To find out whether people are really using AI, read is your team really using AI; if one group is being left behind, read what to do when part of your team is left behind on AI.
What does a passing relay look like in practice?
Take a hypothetical 25-person logistics firm that sends its operations manager on a two-day AI course. In the version that fails, she returns, shares slides at the Monday meeting, and nobody tries anything. Three months later two people use AI regularly, and both were using it before.
In the version that passes, the owner blocks two hours a week in the manager's diary for six weeks before she leaves. Each team gets one task of its own: delivery exception emails, shift handover notes, quote replies. Everyone tries theirs inside seven days, and four weeks later the owner asks two people who did not go to show a result and says what a good one looks like. If the two cannot, the owner knows the relay dropped and where. In one version the firm bought a skill for one person. In the other, it bought one for the business.
What should you do this week?
- List who has been on AI training in the last 12 months and who they taught afterward. If the second list is empty, you have found the gap.
- Run the Relay Test on any seat you are about to book or have recently booked.
- Write the teacher's hours into their calendar before the course, not after. As a suggestion, an hour or two a week for the first six weeks is workable; adjust once you see who has started.
- Choose one real task per colleague and write it next to their name.
- Put the month check in the diary and ask the Confidence Test of the people at the end of the chain.
- If you cannot say yes to all three questions, choose the whole-team or short-module route for the next round. How to get employees to use AI covers the hesitant ones.
Where to go from here
If your team is learning AI one person at a time and nobody can say how well it is spreading, that is the problem my AI training for teams is built around: the whole team, on their own jobs, with time to practice. If you want to know where your business stands before you book anything, try the Business AI Readiness Scorecard. And if you are planning an event for a business audience, I have delivered more than 500 keynotes in 30 countries across 5 continents; the business keynotes page has the details.
Sources and further reading
- Department for Science, Innovation and Technology (with Ipsos), Evaluation of the Flexible AI Upskilling Fund, GOV.UK, published 28 January 2026. UK, baseline and process phase, November 2024 to June 2025; employer survey n=73, employee survey about 105, self-selected.
- Department for Work and Pensions and Skills England, Employer guide: What works for AI upskilling in the UK, GOV.UK, "Skills for AI" published 10 June 2026, HTML guide added 27 July 2026.
- Gask, Coupe and Green, An evaluation of the implementation of cascade training for suicide prevention during the "Choose Life" initiative in Scotland, BMC Health Services Research, 2019. Scotland, 2008 to 2011, qualitative; a different field, used for the mechanism only.
- Sanchez-Moyano, Holmes and Simms, AI Adoption Among Small Businesses: Qualitative Insights from the Small Business Credit Survey, Federal Reserve Bank of San Francisco, 1 September 2026. US, 2024 Small Business Credit Survey.
- On this site: AI training for small business teams, what AI training does your team need, how to train staff to catch AI mistakes at work.
This article is general information, not legal or financial advice.
Dan Fitzpatrick is a keynote speaker, Forbes contributor and five-times bestselling author who has trained more than 150,000 educators in 30+ countries. About Dan Fitzpatrick.
