Somewhere in your organization there is a number about AI, and it is going up. More staff signed in this month than last. More classes used the tool. The seat count rose. Whoever gathered that number did careful work, and it is almost certainly true.
It is also almost useless as a measure of progress, because it would have gone up anyway.
Here is the short answer, before the argument. You cannot measure AI maturity by counting use. Usage counts what people did. Maturity is what your organization can now do reliably: make a decision about AI, tell everyone what it is, and hold it when the person who made it goes on leave. Those two things move independently, and this year they have been moving in opposite directions. So the way to measure maturity is to stop reporting numbers that only rise, and start reporting the few that can fall.
Why usage numbers rise whether you are leading this well or badly
Usage rises because AI is useful and immediately available, not because your leadership is working. Three pieces of research published in the last twelve months make the split visible.
Coursera's AI in Higher Education Report, published on March 11, 2026, surveyed 4,200 students and educators across the United Kingdom, the United States, Mexico, India and Saudi Arabia. Over 95 percent said they use AI tools in an educational setting. In the same survey, 26 percent of educators said their university has a policy governing AI use, and 25 percent believed they and their peers have the skills to use AI to their advantage.
Jobs for the Future published a survey on March 18, 2026, run by AudienceNet with 3,020 US respondents aged 16 and over and fielded between November 28 and December 8, 2025. Two of its numbers went up sharply: 69 percent said AI tools are incorporated into their lessons or training, up from 57 percent, and 69 percent said they had received AI training from their institution, up from 47 percent in 2024. One number more than halved. The share who reported feeling more supported in their learning through AI-assisted resources fell to 13 percent, from 29 percent the year before. Asked where they get their information about AI, 23 percent named their school or training program. Social media got 48 percent.
In the US school system the same shape appears in RAND's survey panels, reported by Christopher Doss and colleagues on September 30, 2025 from nationally representative samples surveyed in spring 2025. Fifty-three percent of English language arts, math and science teachers and 54 percent of students had used AI for school, each up more than fifteen percentage points in one to two years. Meanwhile 45 percent of principals reported having school or district policies or guidance, and 34 percent of teachers reported a policy covering AI and academic integrity.
Read those pairs as one organization improving at two different speeds and you will conclude you are behind on policy. I read them differently, and this is interpretation rather than a finding: usage and maturity are not slow and fast versions of the same thing. They are different quantities, and a usage figure carries almost no information about the other one.
The JFF pair is the sharpest case. Two input numbers rose steeply and the outcome number collapsed. Treat that carefully, because these are self-reports from different samples in different years and "feeling more supported" is a perception rather than an outcome. Even allowing for all of it, the direction matters. More use and more training coincided with fewer people feeling helped. A leadership team looking only at the first two numbers had a good year.
There is a harder version of this. Usage can rise because an organization is immature. When the approved route is slow or unclear, people find their own, which is why unapproved staff use of AI produces a healthy-looking usage curve and a governance problem at the same time. The number cannot tell the two apart.
The four things that get called adoption
Most disagreements about how far along an organization is are really four different words being used as one. Separate them and the argument usually dissolves.
Access. The tool is available. Someone bought or enabled it. This is a procurement fact.
Usage. A person opened it and did something. This is an activity fact, and it is the only one of the four that is easy to collect, which is exactly why it dominates every report.
Practice. The work is genuinely done differently now, and it would still be done that way next month if nobody mentioned AI again. This is the level my Transfer Test is built to check.
Maturity. The organization can hold its decisions about that practice. My working definition has not changed since I first wrote it: being AI ready means your organisation can make, communicate and hold good decisions about AI, consistently, without depending on any one person. Maturity is that state, observed rather than intended.
The rungs run in that order and you cannot skip one. And here is the rule I would ask a leadership team to write at the top of any self-assessment: your maturity is the highest rung you can show evidence for, not the highest rung you can describe.
Usage tells you the tool arrived. Maturity tells you the organization changed.
The Maturity Test
The Maturity Test is three questions I ask of any AI number a leadership team is about to report as progress: Could this number rise while the work itself stayed exactly the same? Could it rise because something is wrong? And if it fell, whose decision would change? A number that survives all three is a measure of maturity. A number that fails any of them is a measure of activity, and activity rises in a well-led organization and a badly led one alike.
This is my suggested filter, not a validated instrument. Its value is that it takes about a minute and it can be applied by anyone in the room to a number already on the screen.
Could this number rise while the work itself stayed exactly the same?
If yes, the number is measuring curiosity, not change. Sign-ins, license take-up, prompts sent, staff who attended the training, teachers who say they have "tried" AI: every one of these can double in a term in which not a single task is done differently. A strong number fails this question, which is to say it cannot rise unless something about the work moved.
Could it rise because something is wrong?
This is the question almost nobody asks, and it is the one that catches usage figures. Heavy use of a general chatbot can mean staff are working faster. It can also mean your approved system is unusable, or that nobody has told anyone what they are allowed to put into it. A figure with two opposite explanations is not evidence for either. A strong number has one reading, and a rise in it is unambiguously good news.
If it fell, whose decision would change?
Ask the person who reports the number what they would do differently if it halved next term. If the honest answer is "nothing", you are looking at a display, not a measure. This is the test's cheapest question and the one that clears out the most material. A number that can only go up is not a measure. It is a scoreboard for a game nobody is refereeing.
Four numbers that pass it
Pick two of these, not all four, and report them next to the usage figure rather than instead of it.
Live replacements. Not what has been added, but what is currently not being done the old way because AI does it: the task, the date it stopped, and the person who signed that off. It can fall, because things quietly restart. A rise means the work moved.
Reported errors. How many times this term did a named person hear that AI got something wrong? Zero is the worst possible reading, not the best. A fall means people have stopped telling you, which is the failure you most want early warning of, and it is the reason this number belongs in front of a board.
Answer consistency. Put the three readiness questions to one senior leader, one middle leader and one person on the front line, separately: What is AI for here? What may I do with it? Who decides when it goes wrong? Count how many of the three give you the same answer. It falls when people leave, when guidance ages, and when a decision was announced rather than communicated.
Single-person dependencies. How many of your working AI uses would stop within a month if one named person left? Every organization I work with can name at least one. A fall in this number is the clearest evidence of maturity there is, and it needs no new data: you already know who the people are.
Two of those take an afternoon. None of them require a platform, a dashboard or a vendor.
The Leadership Question
The question I would put to a leadership team is not "how mature are we?" It is: what is the last decision about AI we made that required us to give something up?
I ask it because of the shape of the answers. A team with a usage report and no decisions will tell me about training delivered, tools enabled, enthusiasm among staff, a working group that meets. All true, all additive, nothing surrendered. A team further along names something it turned down, stopped, or took off somebody's plate, and can tell me the week it happened.
Sitting on both sides of that report helps. As a Director of Digital Strategy in further education, part of my job was producing the numbers that travelled up the building, and the numbers easiest to produce were always the ones about activity. As a school trustee I have since read plenty of reports written by someone in that position. The tell is always proportion: pages on activity, a paragraph on what changed.
Across the schools, trusts and districts I work with, the teams who score themselves lowest on readiness are usually the ones furthest along. They have tried to gather the evidence and discovered what they cannot show. The confident self-ratings tend to come from teams who have had good conversations and have never had to produce a document.
Then are maturity models a waste of time?
No, and the better ones already ask for exactly what I am arguing for. The CoSN and Council of the Great City Schools K-12 Gen AI Maturity Tool, announced on April 9, 2024, rates districts across six domains, including executive leadership, data, security and legal risk, on a scale from emerging to developing to mature. Its own instructions ask districts to supply policy documents, implementation plans, training links and meeting notes to justify a rating, and state plainly that conversations or discussions about something do not constitute evidence.
That is a well-designed instrument. The problem is not the model but what gets fed into it. A team with a usage dashboard and no documents will still rate itself "developing", because it has had the conversations and remembers them as progress. The model asks for evidence; the self-assessment supplies recollection.
Usage numbers have two honest jobs, and it is worth naming them so nobody throws the data away. The first is a baseline, taken once, before you start. The second is a distribution check: not how much use there is, but where there is none. RAND's finding that over 80 percent of students said their teachers did not explicitly teach them how to use AI is a usage figure doing real work, because it is pointing at an absence. Use usage to find the gaps. Do not use it to score the progress.
What to do this term
Take the last AI number your leadership team reported upward. Run the three questions on it. Most numbers fail the first, and almost all fail the third.
Then do three things. Replace it with one number that can fall, and say out loud what decision a fall would change. Ask your three readiness questions at three levels of the organization this month and write down the spread. And name the single-person dependencies, because that list is the difference between a practice and a person.
None of this makes you more mature. It makes your position visible, which is the only honest place a plan can start from.
If your leadership team wants a candid read on where it actually stands rather than a usage report, that is the work behind my School AI Readiness Scorecard and the readiness sessions I run with senior teams.
Sources and further reading
- "AI in Higher Education Report", Coursera, March 11, 2026. Link
- "AI Usage in Education is Growing, But Gaps in Guidance Persist", Jobs for the Future, March 18, 2026, survey conducted by AudienceNet. Link
- Christopher J. Doss and colleagues, "AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels", RAND, September 30, 2025. Link
- "CoSN/CGCS K-12 Gen AI Maturity Tool", CoSN and the Council of the Great City Schools, announced April 9, 2024. Link
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. More about Dan.
