Two different questions wearing the same name
Search for an AI maturity model and you will find a dozen frameworks that broadly agree with each other. The shape is consistent: five stages running from something like ad hoc, through experimenting and operationalising, to transforming and optimising, scored across pillars such as strategy, data, technology, governance and operating model.
These are good instruments and they are not what most people think they are. They measure an organisation's ability to build, deploy, govern and scale AI. They answer a platform question: can this company put a model into production safely and repeatedly.
That is a real question with a real owner, usually a CIO or a head of data. It is simply not the question an AI transformation lead is holding. Being told the company sits at level three changes nothing about which team to coach on Monday, which candidate to hire on Thursday, or which of the five things you get asked about is actually broken.
- An organisational maturity model asks: can we build and run AI systems.
- A people maturity model asks: can the humans we employ actually work with the AI we have already bought.
- A company can be strong on the first and weak on the second. That combination is the most expensive one, because the tools are in place and the return is not.
The five levels of a person
Scored against the work someone actually does, rather than against what they know about AI in general, a person sits at one of five levels. They are deliberately behavioural: each one describes what you would see, not what someone would claim.
- 1. Absent. AI is not part of the work at all.
- 2. Experimental. They have tried it. Nothing stuck, and the week looks the same as it did before.
- 3. Functional. They reach for it on certain tasks, but not by default, and usually not for the hardest parts of the job.
- 4. Integrated. AI is simply part of how the work gets done. Removing it would change the shape of their week.
- 5. Multiplier. They build things the rest of the team then uses, so their capability shows up in other people's output.
The distance between two and three is where most of the population sits and where most training is aimed. The distance between three and four is where the return actually lives, because that is the point at which the work changes rather than getting marginally faster. And level five is the only one that compounds, since it shows up in colleagues who never attended anything.
The ladder has to be read per role
Integrated does not describe the same behaviour in two different jobs. For someone in sales it might mean AI-generated account research is standard and verification before it reaches a customer is standard with it. For a marketer it might mean drafts start from AI and no claim is published without a check. For an operations analyst it might mean exceptions in automated workflows get caught rather than passed through.
So a company-wide level is an average of people being measured against a bar that does not fit any of them. It produces a number that is stable, comparable and describes nobody, which is the same failure mode as a single effectiveness score. The argument in full is in whether your team is actually using AI well.
The move almost nobody makes: turn the level into a bar
Most maturity assessments end as a report. Somebody presents a number, everybody agrees it should be higher, and the document is not opened again. The assessment was run as a measurement rather than as a decision, which is why it changes nothing.
There is one move that makes the whole exercise pay for itself, and it takes a sentence. Find the level your strongest team already holds. Make that the line. Nobody joins below it.
That converts a score into a standard, and a standard is a thing an organisation can actually operate. It also means the bar is real rather than aspirational, because it was set by a team inside your own company that already clears it. Nobody can argue that it is unrealistic.
Where the bar gets used
A standard that only exists in a report is not a standard. It has to attach to moments that already happen, and there are three:
- Onboarding. Every new joiner runs the same read in their first week, so they arrive knowing where the bar sits and what it takes to reach it. This is the one that compounds, because it means the average rises with every hire rather than drifting.
- Hiring. A candidate runs the same conversation before the offer. You see where they actually land rather than what a CV claims, and an AI-written CV has made that distinction considerably more important than it was two years ago.
- Moving people. Promoting someone into your strongest team is exactly the moment to check them against the line that team holds, rather than discovering the gap three months later.
Note that onboarding comes first. A bar used at hiring and not at onboarding filters the people you interview and does nothing for the people you already employ, which is the larger number in every company.
What this does not solve on its own
A ladder is a measurement, and a measurement is not a plan. Knowing a team sits at level two tells you the gap exists; it does not tell you which of the capabilities underneath is missing, and those fail independently. Someone stuck at two because they cannot direct a tool needs something completely different from someone stuck at two because they do not trust the output.
This is also the practical reason the level has to sit on top of dimension-level scores rather than replacing them. The level is what you report. The dimensions are what you act on.
And the gap is usually larger than leaders expect. Prosci's research across 1,107 participants found user proficiency is the single largest challenge in AI adoption, named by 38 percent of respondents, more than double technical integration at 16 percent.
How to run it on one team
- Define what each level looks like for each role in that team before anyone is assessed. If you cannot describe integrated for a specific job, you cannot score it.
- Read every person against their own role's definitions, on the dimensions, and derive the level rather than asking for it.
- Find the level your strongest team already holds and write it down as the bar.
- Attach it to onboarding first, then hiring, then internal moves.
- Re-run it on the same team at ninety days. Movement between levels is the only evidence that any of the enablement worked.
Frequently asked
What is an AI maturity model? Most published AI maturity models score an organisation's ability to build, deploy, govern and scale AI, usually across five stages running from ad hoc to optimising and across pillars such as strategy, data, technology and governance. They answer a platform question: can this company put AI into production safely and repeatedly. They do not measure whether the people in that company can work with the AI it has already bought.
What are the five levels of individual AI maturity? Absent, where AI is not part of the work at all. Experimental, where someone has tried it and nothing stuck. Functional, where they reach for it on certain tasks but not by default. Integrated, where AI is simply part of how the work gets done and removing it would change the shape of the week. And Multiplier, where they build things the rest of the team then uses, so their capability shows up in other people's output.
Why does AI maturity have to be measured per role? Because integrated does not describe the same behaviour in two jobs. In sales it might mean AI-generated research is standard and verified before it reaches a customer; in operations it might mean catching the exceptions an automated workflow got wrong. Scored against one company-wide bar, people with unrelated gaps produce the same number, and the number describes none of them.
How do you turn an AI maturity score into something useful? Find the level your strongest team already holds and make it the line nobody joins below. That converts a score into a standard, and because the bar was set by a team inside your own company that already clears it, nobody can argue it is unrealistic. Then attach it to onboarding first, then hiring, then internal moves. A standard that only exists in a report is not a standard.
What is the difference between AI maturity and AI readiness? Readiness usually describes whether an organisation is prepared to begin: tooling, data, governance and policy in place. Maturity describes how far it has progressed once it has. Both are commonly measured at the organisational level. Neither tells you whether a specific person can do their specific job better with the AI licences they already hold, which is a separate measurement taken from the work itself.