The usage gap is the distance between how much AI leaders think their people use and how much they actually use. It runs one way. The measured number is always the higher one.
Out of every hundred people in your company, how many use AI for work each week? Give the number you would give in a meeting, before anyone has looked it up. Everything on this page is about the distance between that number and the measured one, and about what that distance costs you.
Upstairs the plan for next quarterDownstairs what already happens on a Thursday
First, a question
What would you have said?
Out of every hundred people in your company, how many use AI for work at least weekly? Move the slider to the number you would have given in a meeting last week, then look at what was measured. The size of the difference is the subject of this page.
74out of 100 frontline employees are regular AI users. Two years ago that was more than twenty points lower.1
13out of 100 use it for a meaningful part of their daily work. Leaders estimate four.2
42out of 100 regular users save at least a full working day per week.1
Two different questions, so two different numbers. Weekly use is high and rising. Daily use on a meaningful part of the job is lower, and that is where leaders are furthest off.
Out in the wild
Two versions of the same company, in the same week.
Neither side is hiding anything here. The left column is written by people doing their best with the information they have. The right column is what the information would have said.
What the plan saysWhat is already happening
We start with a pilot in one team, and evaluate after the summer.
Three teams have been using it daily since February. Each one does it differently, and none of them wrote it down.
First a basic training so everyone gets on the same page.
Four people in the room could give that training. Two of them have stopped mentioning it, because the last time they did the conversation turned into a risk discussion.
We will measure adoption in six months.
The time is already being saved. Two thirds of people get no guidance on what to do with it, so it goes into the day and disappears.1
The hard part will be getting people on board.
They are on board. What they are waiting for is permission and a decision, which is a different problem with a different fix.
I have sat in the meeting where the left column is presented while three people in the room are living the right one. Nobody corrects it. Correcting it means explaining what you have been doing since February, and that conversation has a cost attached that the plan does not.
Four people at the same table, asked separately what they actually do:
Team lead, operations“I rewrite every client mail through it. Has done for a year. It is not a secret, it just never came up.”
Account manager“I use it to prepare for calls. If I said that out loud somebody would ask whether the client knows.”
Analyst“Two days of work went to two hours. I filled the rest with more of the same work. Nobody asked me to.”
Junior developer“Everyone here uses it. We just don't talk about it in the standup.”
Composite lines from what people say once you ask them one at a time. The pattern is consistent enough that you will recognise at least one.
What it is
This looks like an adoption problem. It is a map problem.
You saw the number at the top. Adoption was the open question two years ago and it is close to settled now, with three quarters of frontline employees using AI regularly.1 The behaviour has moved. What has not moved is the picture leadership is working from, and that picture is what budgets, training and hiring plans are built on.
So the cost lands somewhere other than where people look for it. You are making decisions for a company that no longer exists, and the plans still read as reasonable on paper, which is what makes this one hard to spot from the inside.
Nobody in that meeting is lying. They are answering a question about a company that stopped existing about eight months ago.Paul Musters, founder of emaho
There is a version of this that is about hiding, and it is worth separating. Some people keep their AI use quiet on purpose, and that has its own reasons and its own fix. The usage gap is broader. It also covers the people who would tell you happily, if anyone had asked.
The measured part
The workforce moved. The guidance did not.
BCG asked 11,749 workers across fourteen markets this year.1 Two figures out of that survey matter more than the rest, and they sit next to each other.
The time is being saved. Most of it leaks straight back out, because nobody said what it was for.1
A day a week, for four in ten regular users. That is the largest unmanaged asset in most companies right now, and two thirds of the people holding it have been told nothing about what to do with it. It goes into more of the same work, or into the gaps in the day, and by the quarter close there is nothing to point at.
47%
now spend more time directing AI than doing the work themselves
That last one is the argument for doing anything at all about this. The lever sits in whether people know what the company is trying to do with the time and the capability they already have.
Employees don't push back on AI intensity; they thrive when the strategy is clear, the direction is real, and the message reaches them.Sylvain Duranton, global leader of BCG X1
The mechanism
Three reasons the map goes out of date.
Nobody is asked, so nobody answers
Most leadership teams have never put the question to their people directly. The picture comes from what surfaces in meetings, and what surfaces in meetings is whatever somebody felt safe mentioning. That is a sample, and not a good one.
The people furthest ahead are the least visible
Whoever has rebuilt their week around it has the least reason to bring it up. There is no reward for it and a real chance of a discussion about risk. So the person with the most useful information in the building says the least, which skews the map in exactly the wrong direction.
The person who could redraw your map for you in ten minutes is usually the one who has learned not to raise it.
Paul Musters
A number from March feels current in September
This one is almost mechanical. Adoption is moving in tens of points per year, and a survey you ran two quarters ago describes a company that has moved on since. Half a year is a long time now. Most planning cycles were built when it was not.
Put a level on the whole company
Three minutes and six questions, answered anonymously. The level, the cost, and the step after it.
Four questions, anonymous, before your next AI decision.
This is deliberately not a survey project. One form, four questions, anonymous, closed after a week. If your next decision about AI is worth making, it is worth making on this instead of on the picture in your head.
Ask these four. In this order.
How often do you use AI for work: daily, weekly, sometimes, never.
What do you use it for. Name the task, not the tool.
If it saves you time, roughly how much per week, and where does that time go now.
What would you use it for if nobody minded.
Answer four is the one people skip and the one worth reading twice. It tells you where the permission is missing, which is usually cheaper to fix than anything on your roadmap.
Write your own estimate down before you open the results. Treat it as a calibration. The size of your error is the thing you are actually measuring, and it tells you how much of the rest of your plan to trust.
Level by level
A gap this size is a Level 2 symptom.
01Campfire60%
02Wild West25%Most companies are here
03Blueprint10%
04Engine4%
05Ecosystem1%
Share of companies per level. The five AI Culture Levels we use with clients, from everyone doing their own thing to AI being part of how the company runs.4
In the five AI Culture Levels we use with clients, Level 2 is daily individual use with no shared standard. Everybody has found something that works and nobody does it the same way, so there is nothing central to look at and the picture upstairs stays a guess. About a quarter of organisations are there.
The step up is writing down how the work is actually done, which is only possible once you know how it is being done. That is why this page starts with asking rather than with a policy. You cannot document a way of working you have not seen.
What a measurement gives you that a survey does not
An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.
An anonymous survey closes the gap once. It tells you where you were in the week you asked, and six months later you are guessing again. It also gives you a total, when what you need is a distribution: who is three levels ahead, who has not started, and which team is carrying the whole average on its own.
The unit is the person. How somebody thinks and works, and how far along they are with AI. You get a level per person and a picture per team, and it stays current because you can run it again. The people at the top of that distribution are the ones to build the standard with. They already have one, they just never wrote it down.
About emaho
emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.
Fifteen minutes per person. No credit card, no strings.
The answer
Four steps, and the first one is this week.
Ask, anonymously, before the next decision. Four questions, one week, no project plan. Then write your own estimate next to the result and show both to your leadership team, because the error is the more useful of the two numbers and it will do more for the conversation than the data will.
Second, decide what the saved time is for. Say it out loud and put it somewhere people can see it. Two thirds of people are given nothing here, so almost any clear answer beats the current situation.1 It can be simple: we want the extra day going into client conversations this quarter. What kills the value is silence, because silence gets filled with more of the same work.
Third, go and find the three or four people who are furthest ahead, and ask them to write down how they actually do it. Ask for one page each. A training is the wrong shape for this. Those people exist in every company I walk into, they are rarely the ones you expect, and they will do it gladly the moment somebody in charge treats it as useful rather than as a risk to be managed.
And then repeat it in a quarter. Adoption moved twenty points in two years, so a number from March is already describing a different company by September. If you only do this once you will be back where you started by spring, holding a map you trust more than it deserves.
Send me your estimate and your result
Two numbers, that is all I need. I will tell you what that gap usually means at your size and which of the four steps I would do first. No pitch. An answer, from me, usually the same week.
Two companies on one payroll, and only one of them is using this.
Start with the money, because it is the easiest to see. Training budgets aimed at beginners, licences bought for the wrong group, a pilot for something three teams already do. None of that is wasted because the plan was bad. It is wasted because the plan was answering a question about last year.
Then the day a week that four in ten regular users are already saving.1 Nobody is stealing it. It goes into more of the same work, which feels productive and shows up nowhere, and by the time somebody asks what AI has delivered there is no honest answer available even though the time is real.
The slowest cost is what it does to the people who moved first. They took a risk, it worked, and the organisation carried on as though it had not happened. After the third planning cycle that ignores them they stop offering. You still have them on the payroll and you have lost the thing that made them worth listening to.
Questions
The usage gap, in questions
How many employees actually use AI at work?
Most of them, and more than leadership tends to think. BCG's 2026 AI at Work survey of 11,749 workers found 74% of frontline employees are now regular AI users, up more than twenty percentage points in two years. For daily use on a meaningful part of the job the figure is lower, around 13%, against the 4% leaders estimate in McKinsey's January 2025 research.
What is the AI usage gap?
It is the distance between how much AI leaders believe their people use and how much they actually use. It runs in one direction: the measured number is always the higher one. The practical problem is not the behaviour, it is that budgets, training plans and roadmaps get built on the lower number.
Why do leaders underestimate how much their team uses AI?
Because almost nobody asks directly, so the picture comes from whatever surfaces in meetings. The people furthest ahead have the least reason to raise it, since there is no reward for it and a fair chance of a risk conversation. That means the person with the most useful information says the least.
How do you find out how many of your people really use AI every day?
Ask the people rather than the licence dashboard, and do it anonymously. emaho reads organisations on five AI Culture Levels with a scan of six questions that takes three minutes, and gives you the level your organisation is on, what it costs you there, and what changes one step up. Most leadership teams guess a level too high.
How do I find out how much my team really uses AI?
Ask anonymously, four questions, one week, before your next AI decision. How often do you use it, what for, how much time does it save and where does that time go, and what would you use it for if nobody minded. Write down your own estimate before you open the results, because the size of your error tells you how much of your current plan to trust.
Is the AI usage gap the same as shadow AI?
They overlap and they are not the same. Shadow AI is the deliberate part: people using tools nobody approved, often in personal accounts, and staying quiet about it. The usage gap is broader and includes everyone who would have told you happily if anyone had asked. One is a trust problem, the other is a measurement problem.
What happens to the time AI saves employees?
In most companies it disappears. BCG found that 42% of regular frontline users save at least a full working day per week, while 66% receive limited or no guidance on what to do with that time and more than half do not redirect it into strategic work. It goes back into more of the same work, which feels productive and shows up nowhere.
How do you know which people are actually ahead with AI and which are stuck?
A usage report cannot tell the difference between somebody running whole workflows and somebody shortening emails. emaho measures one Operating Profile per person, personality type and AI level in a single profile, so you get the distribution instead of an average. That is what tells you where the next euro should go.
Our AI project shows no results. Is this why?
Often, yes. If the time saved was never pointed at anything, the value leaks out of the organisation before it reaches a number anyone reports on. BCG found that a clear strategy lifts measurable business impact by around 25 percentage points, while better tools alone move it about five.
How often should we measure AI adoption?
Every quarter, and keep it small. Frontline adoption moved more than twenty points in two years, so a figure from six months ago describes a company that has changed since. A short repeatable question set beats a large survey project you will only run once.
How does the usage gap relate to AI maturity levels?
A large gap is a Level 2 symptom in the five emaho AI Culture Levels, where use is daily and individual with no shared standard, so there is nothing central for leadership to look at. About a quarter of organisations sit there. Moving to Level 3 means writing down how the work is actually done, which is only possible once you have asked.
What connects to it
The gap has a few close relatives.
Once you have asked and the answer surprised you, three other questions arrive quickly.
The four questions give you this week's picture. A profile per person gives you the distribution, and it stays current, which is the part a survey cannot do. Start with your own.
Write your guess down before you measureThe distance between the guess and the count is the part worth discussing.
Count once, per personFifteen minutes each, and you have a distribution rather than a licence total.
Read the spread, not the averageThe average hides the team that is three levels ahead and the one that has not started.
Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.
Boston Consulting Group, AI at Work, fourth annual edition, press release of 3 June 2026. Global survey of 11,749 workers across 14 markets. Source of: 74% of frontline employees being regular AI users, up more than 20 percentage points over the previous two years; 72% saying AI has considerably changed skills expectations in their role; 47% spending more time managing and directing AI than doing the work itself; 67% of regular users reporting improved job satisfaction alongside 41% reporting increased cognitive load; 42% of regular frontline users saving at least a full working day per week; 66% receiving limited or no guidance on what to do with that time, with more than half not redirecting it into strategic work; and clear strategy lifting measurable business impact by 25 percentage points against roughly 5 points for better tools. Quotes from Vinciane Beauchene and Sylvain Duranton come from the same release. BCG defines frontline employees as individual white-collar employees without managerial responsibilities.
McKinsey, January 2025. Leaders estimate that about 4% of employees use generative AI for a meaningful part of their daily work, while employees themselves report 13%.
Ivanti, Technology at Work, 2025. 42% of office workers use generative AI at work and one in three of them keeps it quiet.
emaho AI Culture Levels. Level 2, Wild West, is daily individual use with no shared standard, and holds about a quarter of organisations. Calibrated against BCG 2025 and McKinsey 2025.
Numbers are quoted as published. The four voices in the recognition block are composites of what people say when asked one at a time, not quotations from named individuals.
