The people left behind on AI in a small business are the roles missing one of four things: a tool they can use, permission to use it, protected time to learn it, or a named person to ask. Check by role, not by person. The Access Grid is a one-page table that does it in an afternoon.
Most owners discover the gap by accident. Two people use AI every day, one mentions it in a meeting, and everyone else goes quiet. The quiet is the finding.
Is the AI gap in your team a skills problem or an access problem?
It is more often an access problem than owners expect, and access is the part an owner can change in a week. Three recent sources point the same way, though none of them measured a small team by role, so treat the match as a pattern and not a proof.
PwC's 2026 Global Workforce Hopes and Fears Survey, published on September 29, 2026, asked 49,364 workers in 48 countries and regions in May and June 2026. Some 64% had used AI at work in the past 12 months, up 10 points, and daily generative AI use rose from 14% to 22%. Yet only 51% said they had the learning and development resources they need, down from 59% a year earlier. PwC's largest group, the 56% it calls the "Engine Room", are workers without scarce skills and without much AI experience. Only about two in five of them said they could get the learning they need.
Gallup's workplace AI study surveyed 23,717 employed US adults from February 4 to 19, 2026. Among employees whose organizations make AI available, 67% of leaders use it a few times a week or more, against 52% of managers and 46% of individual contributors. Employees who strongly agree their manager actively supports AI use are far more likely to be frequent users: 78%, against 44% of those who do not strongly agree. For clear AI policies the figures are 68% and 47%. Having the tool, in Gallup's reading, does not guarantee use.
The UK Department for Work and Pensions published its Skills for AI evidence summary on July 27, 2026, drawing on 23 workshops with around 150 organizations, 10 case studies and a survey of 536 UK employers. It found that staff often learn by trial and error, peer support and online videos, which helps people start but can produce "uneven and risky practice". The main barrier it names is capacity, not interest: limited time, staff pressure and training that is not linked to real roles.
Official UK statistics show why the role matters. In the Office for National Statistics' Wave 165 survey, fielded from September 21 to October 4, 2026 among UK private sector businesses, 38% of AI-using businesses with 10 or more employees reported effects on administrative and clerical roles, the role group named most often. AI reaches some jobs first, and the roles it reaches first need the same support as the rest.
Here is my reading. Interest is rarely the missing ingredient. What is missing is something a manager controls: a tool, a clear yes, an hour, a name to ask. The roles that lack them are often the ones the owner sees least of.
What is the Access Grid?
The Access Grid is my suggested way of checking who in a team is being left behind on AI. It is a way of thinking, not established research.
The Access Grid is a one-page table with a row for every role in the business and four columns: Tool, Permission, Time and Person. For each cell the owner writes Yes, Partly or No, based on what people in that role would say, not what the policy says. Any role with two or more cells that are not Yes is a role being left behind, and the first fix goes there before anyone gets advanced training.
A role is a group of people who do the same kind of work: site fitters, front-of-house staff, the accounts team, the sales desk. Rate roles, not individuals. You are looking for a gap in the system, and a gap in the system is what you can fix.
Copy this table into a document or a spreadsheet:
| Role (headcount) | Tool | Permission | Time | Person |
|---|---|---|---|---|
| Role 1 | Yes / Partly / No | Yes / Partly / No | Yes / Partly / No | Yes / Partly / No |
| Role 2 | ||||
| Role 3 | ||||
| Role 4 | ||||
| Role 5 |
Rate each column by asking one question of the role:
| Column | The question | Yes | Partly | No |
|---|---|---|---|---|
| Tool | Can they open an approved AI tool today, without asking a favor? | Yes, on their own login and device | Shared login, or only on the office computer | No approved tool |
| Permission | Can they say in one sentence what they may and may not use AI for? | They can say it | They know rules exist | They are guessing |
| Time | Do they have protected time to practice on their own work? | Time is in the diary | "When it is quiet" | None |
| Person | Can they name who they would ask when stuck? | They name a person | They name "whoever" or "the internet" | Nobody |
How do you fill in the Grid?
Ask the role, not the policy. A policy that exists in a shared folder is not the same as a permission a fitter can recite on a wet Tuesday morning.
- List the roles. Five to eight is enough for a business of 20 to 50 people. Write the headcount beside each.
- Pick two people per role and give each ten minutes in private. Ask three things: show me the last time you used AI for work; what stopped you the last time you wanted to; who would you ask if it went wrong?
- Rate each cell using the table above. When the two people disagree, take the lower rating. The person who found the workaround is not the one you are checking on.
- Count the non-Yes cells in each row. Two or more means the role is being left behind. One means it is at the edge.
- Add up the headcount of the roles left behind. That share of your team is your honest answer to "are we using AI?"
What does a filled-in Grid look like?
Take a hypothetical 24-person kitchen design and fitting firm. The directors and the office use AI to draft emails and quotes, and they have talked about it enough that it feels as if the whole firm is on board. The Grid says otherwise.
| Role (headcount) | Tool | Permission | Time | Person | Non-Yes cells |
|---|---|---|---|---|---|
| Directors (2) | Yes | Yes | Partly | Yes | 1 |
| Office and accounts (3) | Yes | Yes | Partly | Yes | 1 |
| Designers (4) | Yes | Yes | Partly | Yes | 1 |
| Workshop (5) | No | No | No | No | 4 |
| Site fitters (10) | Partly | No | No | No | 4 |
Fifteen of the 24 people, nearly two in three, sit in roles with four gaps each. The nine people who use AI are the nine who talk to the owner. That is the same trap I describe in the depth check for your team's AI use: a few enthusiasts shape what the owner believes, and the quiet roles never appear in the numbers.
Look at what the fix would be. The fitters do not need a course. They need a one-page note on what they may type, photograph or send (permission), a designer named as the person to ask on Thursday afternoons (person), a 20-minute slot at the Friday team meeting (time), and the approved tool on the phone they already carry (tool). Permission and person cost a conversation. They come first.
What do you fix first?
Fix Permission first, then Person, then Time, then Tool. The first two cost a conversation, not a project, and they are the two that Gallup's figures on manager support and clear policies point toward.
| Column | The smallest useful fix | Who owns it | By when |
|---|---|---|---|
| Permission | One page, in plain words: what you may use AI for, what you may never paste in, who to tell when it goes wrong. The UK small business AI policy template is a starting point; the shadow AI article explains why a clear yes beats a silent ban. | The owner | This week |
| Person | Name one ask-person per role, someone in the role who already uses AI well. Give them 30 minutes a week for questions. | The role's manager | This week |
| Time | Put a fixed slot in the diary, 20 to 30 minutes, for practice on real work. Protected means nobody can move it for a job. | The role's manager | Within two weeks |
| Tool | Give the role an approved tool on the device it actually uses, with one saved example of a task worth doing. | The owner | Within a month |
Do not start with advanced training for the roles already ahead. It feels productive, and it widens the gap. When you do get to formal training, the one-page training brief sorts what each person is missing, and the Four Rungs tell you why training that skips permission and practice tends to fade.
What I See When I Train Business Teams
I have worked with organizations as different as Google, Grammarly, Glasgow City Council and the North East Local Enterprise Partnership. What follows is a composite of the pattern I see across business teams, not a story about any one of them.
The keen two or three set the tone for what everyone believes. They are usually the people closest to the owner, and they are often the people who needed the least help. The others rarely say "I haven't been shown." They say "it isn't really for my job." That sentence sounds like resistance. Often it is a polite way of saying that nobody has told them they are allowed, nobody has given them an hour, and they do not know whom to ask.
If your team contains skeptics as well as the unsupported, separate them first. Getting skeptical employees to use AI is a different problem with a different fix. The Access Grid finds the people who would use AI if the way were open.
How do you know the gap has closed?
You know it has closed when a person in your most-left-behind role passes 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.
Ask it about six weeks after the fixes, and ask it of the role that had four gaps, not the role that already had one. Then redo the Grid at 90 days. A good result is no role with more than one non-Yes cell. A managers' routine such as the Friday Four keeps the gain from fading, because it gives every role a weekly place to say what it used AI for and what it got wrong.
What should you do this week?
You can finish the first pass in five working days without a project or a budget meeting.
- Monday: List your roles and headcounts. Stop at eight.
- Tuesday: Hold the ten-minute conversations, two people per role, using the three questions above.
- Wednesday: Fill in the Grid and count the headcount in roles with two or more gaps.
- Thursday: Write the permission page and name the ask-person for each role.
- Friday: Put the practice slot in the diary, then add the six-week Confidence Test and the 90-day Grid to the calendar.
Take the Grid to your next leadership meeting before you take any new AI tool. The share of your team that is left behind is a number worth knowing before you spend another month of attention elsewhere.
Where to go next
If your team is using AI unevenly and nobody has shown the quiet roles how to start, that is what my AI training for teams is built for. If you want to see where the whole business stands first, the Business AI Readiness Scorecard takes about five minutes. And if this question belongs in front of your leadership team or members, take a look at my business keynotes.
Sources and further reading
- PwC, "Companies risk losing their most AI-savvy employees while leaving majority behind", press release for the 2026 Global Workforce Hopes and Fears Survey, September 29, 2026 (49,364 workers, 48 countries and regions, fieldwork May and June 2026): pwc.com
- Gallup, "AI in the Workplace: What Separates Adopters and Holdouts", April 12, 2026 (23,717 employed US adults, fieldwork February 4 to 19, 2026): gallup.com
- Department for Work and Pensions with Skills England, "Skills for AI: what works for AI upskilling in the UK", executive summary, July 27, 2026 (UK; 23 workshops, 10 case studies, 536 employer survey responses): gov.uk
- Office for National Statistics, "Business insights and impact on the UK economy", October 8, 2026 (Wave 165, fieldwork September 21 to October 4, 2026; UK private sector businesses; official statistics in development): ons.gov.uk
Dan Fitzpatrick is a keynote speaker, Forbes contributor and five-times bestselling author who has trained more than 150,000 people across 30+ countries. About Dan
