Every person in a small business needs the same floor of AI skills, and each level of responsibility adds one rung on top. The floor is knowing what you may use AI for, using it on your own work, and saying when it gets something wrong. Entry-level staff add using and checking, mid-level staff add owning a workflow, and managers add deciding and answering for the result.
Sources checked November 21, 2026. The Skills England material is guidance for employers in England, and I use it as a framework to borrow, not a standard your business must meet. The survey figures are from UK employers and are older than the tools they describe, so I read them as direction, not as a measurement of today.
Why is it so hard to say which AI skills a role needs?
Because most owners are guessing, and the evidence says so. In an Ipsos survey for the UK government, AI Skills for Life and Work: Employer survey findings (Department for Digital, Culture, Media and Sport, published January 28, 2026), 801 UK employers answered by phone and online between March 19 and June 7, 2024. Sole traders were excluded. Only 11% of all employers said staff had done any AI training in the previous 12 months.
The main barrier to training was not money. "The main barrier is being unsure what training is relevant for the business, cited by 50% of employers," the report says. Lack of time came second at 47% and cost third at 41%. All three figures are shares of all 801 employers.
The skills ratings tell the same story from the other side. Among the 348 employers who had staff working with AI, 61% rated their staff as good at keeping information safe and private, and 45% rated them good at writing clear prompts. That is a self-assessment by employers, not a test of the staff. It still suggests that even where people are using AI, the skills are uneven and nobody has written down which ones matter for which job.
That is the gap this article fills. I am not claiming the survey measured it. The report is about employers in general and does not break the results down by size, so I make no claim about businesses of your size from it.
What does the government's AI skills framework say?
It sorts AI skills into three domains and three job levels, and says plainly that the levels are a guide. Skills England's AI skills tools package (first published October 29, 2025, last updated November 4, 2025) contains an AI Skills Framework, an AI Skills Adoption Pathway Model and an Employer AI Adoption Checklist.
The framework's three domains are technical, responsible and ethical, and non-technical. Its three levels are entry, mid and managerial. Two sentences from the page matter for a small business. The levels are "intended as flexible guides to help identify relevant skills, not as rigid classifications," and "the framework is cumulative," which means higher levels keep the core skills of the lower ones. Its examples include writing prompts and routine digital tasks with AI tools at entry level, and planning AI strategy and responsible use expectations across departments at managerial level.
Here is my reading, and it is only that. Nine cells, three domains by three levels, is a sensible map for a large employer. For a 20-person business it is too many boxes for anyone to keep current, and the page itself has no section for small employers. What survives the trip down to a small team is the cumulative idea: everyone keeps the basics, and responsibility adds to them. The Floor and the Rungs is that idea made small enough to use.
What is the Floor and the Rungs?
It is a way to set AI skills by role with one standard everyone shares and one addition per level. Here is the definition I use.
The Floor and the Rungs is how I suggest a small business sets AI skills by role: one floor that everyone stands on, which is the Confidence Test passed in practice, and one added rung for each level of responsibility. Entry-level staff use and check. Mid-level staff own a workflow. Managers decide and answer for the result.
This is my suggested way of thinking, built on the cumulative design of the Skills England framework. It is not established research, and no study has tested it.
| Level | Added rung | Roughly where it sits in the Skills England domains (my mapping) | The evidence I would accept |
|---|---|---|---|
| Everyone | The floor: knows what AI may be used for, uses it on their own work, says when it is wrong | Responsible and ethical, plus everyday non-technical skill | Three yes answers to the Confidence Test, from the person themselves |
| Entry | Use and check | Technical: prompts and routine tasks. Responsible: checking before anything leaves the building | Two real pieces of work with the checks marked and one mistake caught |
| Mid-level | Own a workflow | Technical and non-technical: a repeatable job, written down | Someone else can run the written steps and get a usable first draft |
| Manager | Decide and answer | Non-technical: setting expectations across the business. Responsible: accountability | A one-page list of approved uses, a named checker for each, and a review date |
What is the floor everyone stands on?
The floor is the Confidence Test, asked of the person doing the work and answered with what they actually did. Here is the wording I use every time.
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.
Used as a floor, it does two jobs. It is the same for the owner and the newest hire, so nobody is excused. And it tests behavior, not knowledge. A person can pass a course and still answer no to the third question, because in their business telling someone about a mistake feels like confessing to one.
The first question needs a written answer to exist. If you have not told people what AI is for here, what they may do with it and who decides when it goes wrong, no one can say yes honestly. That is the Readiness Test, and it comes before any skills sheet.
What does each rung add?
Each rung adds one new responsibility and takes nothing away. People keep the floor at every level.
Entry: use and check. The person uses AI on tasks that are part of their job and checks every name, date, figure and claim against the source before it moves on. The skill to build is checking, not clever prompting. How to teach it is covered in how to train staff to check AI output.
Mid-level: own a workflow. The person takes one repeated job and turns it into steps someone else can follow: what goes in, what never goes in, what the prompt asks for, what a checker looks at. This is where prompting as a job skill belongs. It is a written method for a real job, not a trick.
Manager: decide and answer. The manager decides which jobs AI may touch and answers when something goes wrong. This rung has one test I would not skip. Before accepting "a human checks it" as an answer, apply 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.
A manager who cannot name the checker for each approved use, and say what that person would see, has not reached the rung yet. In an owner-run business the owner holds this rung until they hand it to someone by name.
How do you fill in a Role Skills Sheet?
You fill in one short sheet per role, in about fifteen minutes, with the person who does the job. Copy this and put it in a shared document.
ROLE SKILLS SHEET
Role: Level (entry / mid / manager):
Person: Date written: Review on:
THE FLOOR (ask the person, not their manager)
[ ] I know what I am allowed to use AI for in this role
[ ] I have used it on my own work this week
[ ] If it got something wrong, I would tell someone
THE RUNG FOR THIS LEVEL (write one sentence in plain words)
In this job, I use AI to: ____________________________________
I check it by: _______________________________________________
Never goes into an AI tool from this role: ___________________
(mid and manager only) The workflow I own is: ________________
(manager only) Approved uses / named checker for each: ________
EVIDENCE I WILL ACCEPT (something I can look at, not a certificate)
____________________________________________________________
WHAT WOULD HELP THIS PERSON MOST NEXT: _______________________
Here is a hypothetical example, so you can see the level of detail. Take a hypothetical 30-person building-services firm. It has an office administrator, an estimator, an operations manager and an owner. I have made this firm up for illustration.
| Role | Rung in one sentence | Evidence |
|---|---|---|
| Office administrator (entry) | Uses AI to tidy customer emails and job notes, and checks every name, date and figure against the job file before sending | Two sent messages with the checks marked, and one error she caught |
| Estimator (mid) | Owns the quote-drafting steps: what goes in, what never goes in, what the checker compares against the price list | A second estimator runs the steps cold and gets a usable draft |
| Operations manager (manager) | Decides which jobs AI may touch and names a checker for each | A one-page list with a checker named and a review date |
| Owner (manager) | Holds the rung until the operations manager takes it, and passes the Loop Test on each approved use | The same page, signed off with the date |
Notice what is missing: no tool names, no course titles and no scoring. The sheet describes the work and the proof. If you want to find out what is blocking a particular person once the sheet is filled in, my Gap Test is the next step, and it sorts one person's gap on one job. If the role itself is changing, rewriting the job description is a separate piece of work.
The Mistake I See Most Often
The mistake is a skills list written about the tool instead of the job. "Understands prompts" and "knows the main AI tools" sound like standards, but they cannot be seen, so nobody can say when a person has them. The sheets that hold up describe a piece of real work and the proof that it was done. I also see owners put the floor off until after a training day, when the floor is what makes a training day worth running. Both are patterns across the business teams I train, not a count, and I would rather say that than invent a figure.
I come at this as a former teacher, assistant headteacher and Director of Digital Strategy in further education, which means I spent two decades writing down what a person should be able to do and how you would know. The hard part of AI at work is not the technology. It is helping people use it well, safely and every day. That is what an educator does.
What if my business only has eight people?
Keep the floor and drop the levels. With eight people you usually have one manager rung, which the owner holds, and one working rung, which everyone else shares. Write the floor once, write one sheet for the owner and one for everyone else, and review them together in a single 30-minute meeting. The sheet is for finding out who needs what, not for filing paperwork about a team where everyone already knows each other's work.
What should you do this week?
You can do this in an afternoon, with no software and no data team.
- Pick three roles that already use AI, or are about to. Include one manager.
- Ask three people who are not leaders the Confidence Test. Ask on their own, not in a meeting. Write down the third answer word for word.
- Fill in one Role Skills Sheet per role, with the person who does the job, using the template above.
- For the manager role, apply the Loop Test to every use of AI you have approved. If a check fails, change the check, not the wording.
- Put a review date on the calendar, six to eight weeks away, and look at the evidence column then.
If the third answer to the Confidence Test is no, that is your first finding, and it is about how mistakes are treated, not about AI.
If you want to know where your business stands before you write any sheets, take the free Business AI Readiness Scorecard: 12 questions, about five minutes, with a PDF report and a 90-day plan. If your people are using AI but nobody has shown them what good looks like in their own jobs, that is what my AI training for business teams is built for, and you can talk to me about it.
Skills sheets do not make anyone more capable by themselves. They make it obvious what you are asking of each person, and that is the part owners usually leave unsaid.
Sources and further reading
- AI Skills for Life and Work: Employer survey findings, Ipsos for the Department for Digital, Culture, Media and Sport, published January 28, 2026 (fieldwork March 19 to June 7, 2024; 801 UK employers, sole traders excluded).
- AI skills tools package, Skills England, first published October 29, 2025, last updated November 4, 2025.
Dan Fitzpatrick is The AI Educator: a keynote speaker, Forbes contributor and bestselling author who has trained more than 150,000 people across 30+ countries. More about Dan.



