Your organization does not have an AI maturity level. It has a distribution. The single number on most AI maturity assessments, stage two, stage three, a point on a five-point scale, is comfortable because it is simple and misleading because no enterprise of any size operates at one level of AI maturity. Your engineering function may be running agentic workflows in production while your finance team still uses a chatbot as a faster search bar. Averaging those into one corporate score does not summarize the organization. It deletes the only information a leader could act on: which teams are getting a return, and which are not.
This matters because the gap between adoption and impact is now the central enterprise problem, and that gap lives at the team level. McKinsey's State of AI 2025 found 88 percent of organizations using AI in at least one business function, up from 78 percent a year earlier, yet only about 6 percent qualify as high performers who attribute more than 5 percent of EBIT to AI. Read as a company story those numbers are demoralizing. Read as a team story they are actionable: inside almost every large enterprise there are teams at high-performer level and teams at stage one, and the corporate average hides both. The teams that are working are a playbook. The teams that are stuck are a cost. A single maturity score tells you about neither.
The same uneven distribution shows up across companies, not just inside them. PwC's 2026 Global AI Jobs Barometer, which analyzed more than one billion job postings across 27 countries, found a pronounced superstar effect: the top fifth of the most AI-exposed companies posted average labor productivity growth of 163 percent since 2018, far ahead of the rest of the field. The point is not that a few firms are special. It is that AI returns concentrate, and concentration is exactly what an average erases. What PwC measured across companies is the same shape you are missing inside your own: a handful of teams pulling far ahead, most clustered behind, and a single score reporting neither.
Why does the company-level view persist?
It persists because company-level data is the data that is easy to collect. You can survey the C-suite, count the seats, and tally the tools. Every one of those is a company-level proxy, and every one is a leading indicator at best.
Seats are the clearest trap
A vendor will tell you how many licenses are active. Active licenses are not outcomes. METR's July 2025 randomized controlled trial found experienced developers were 19 percent slower with AI on familiar work while reporting themselves 20 percent faster, a 39-point gap between what people felt and what was measured. If a fully adopted, fully active team can be slower and not know it, then seat counts and self-reported sentiment are not measuring impact. They are measuring the appetite for it.
This is the measurement infrastructure gap, and it is the real reason most enterprises are stuck. They are not stuck because they lack AI. They are stuck because they cannot tell which of their teams are getting a return and which are paying a productivity tax, so they cannot move resources, redesign workflows, or retire tools with any confidence.
Consider what the average actually costs. If a maturity score says stage three, a leader funds the next phase uniformly: more licenses, another training rollout, a broader mandate. But the return is not uniform, so uniform investment pours budget into teams already underwater and starves the teams that found the playbook. The score does not just fail to inform the decision. It actively misdirects it. A distribution would have told the same leader to study the teams pulling ahead and to fix or pause the teams paying a tax. That is the difference between a number that decorates a board deck and a number that changes where the money goes.
What the team-level view requires
Measuring at the team level is harder than scoring a company, and that difficulty is exactly why it is defensible. A company score can be assembled from a survey in an afternoon, which is precisely why it persuades no one who has to act on it. A team-level number has to be earned from the behavioral record, and that cost is what makes it worth trusting. Three things are required.
Behavior, not sentiment
The signal has to come from how work actually moves across the tools a team already uses, not from asking the team how it feels. Sentiment lags behavior by a full survey cycle and overstates gains for tools leadership has endorsed.
A control
A team's before-and-after change is not a measurement, because dozens of other things changed in the same window. The honest unit is a matched comparison: the change in the AI-using team against the change in a comparable team, matched on tenure, role mix, type of work, and tool stack. This is controlled cohort analysis, and it is the difference between a number a CFO can defend and a number a CFO has to apologize for.
A reliability score honest about partial data
Real teams have gaps. The score has to say, in the open, how much of the picture it could see and how much weight the reader should place on the result, rather than presenting every estimate with false precision.
Company score versus team measurement
| Company-level maturity score | Team-level measurement |
|---|---|
| One number on a five-point scale | A distribution across teams |
| Built from C-suite survey and seat counts | Built from behavioral signal in the work tools |
| Averages the best and worst teams together | Names which teams return and which pay a tax |
| Cannot guide a resourcing decision | Tells you where to move budget and redesign work |
| Stable and comfortable | Defensible and actionable |
None of this fits on a five-point corporate scale, and it should not. The maturity slide flatters the organization; the distribution improves it. One is built to be presented, the other to be acted on. The organizations that pull ahead over the next two years will be the ones that stopped asking what stage they are at and started asking which teams are getting a return, by how much, and why. That is the work behind the Levos platform and the AI Impact signal family, and the matched-comparison method is described in the Levos measurement methodology. The question has a number. It just is not the one on the maturity slide.
Frequently asked questions
What is AI maturity measurement? Assessing how well an organization turns AI use into outcomes. The common form is a single company score, which averages away the only actionable signal: which teams return and which do not.
Why is a company-level score misleading? It deletes variation a leader could act on. McKinsey found 88 percent of organizations use AI but only about 6 percent are high performers tying it to real EBIT. Those high performers are teams hidden under the average.
Why are seat counts a poor measure? Active licenses are not outcomes. METR found experienced developers 19 percent slower with AI while feeling 20 percent faster. Seats measure appetite, not impact.
How do you measure AI maturity at the team level? Behavior not sentiment, a matched control (controlled cohort analysis), and a reliability score honest about partial data. None of it fits a five-point scale.
Stop scoring the company, start measuring the teams
Levos is opening early access to a small cohort of enterprise leaders who want to see AI return at the team level, where the playbook and the cost both live, rather than as a single company stage.
