Everyone in your company has heard of ChatGPT. That tells you nothing useful. Here is how to work out where your organization actually is, and what has to change to move.
There is a wide gap between adoption and fluency, and the numbers make it visible. Stanford’s AI Index reports 88% of surveyed organizations now use AI, yet agent deployment sits in the single digits across nearly every business function[3]. IBM’s 2026 CEO study found leaders believe 86% of their employees have the skills to work with AI while reporting that only 25% of the workforce uses it regularly[7]. This guide defines four stages between awareness and fluency, explains why fluency is a multiplication problem rather than a checklist, sets out what actually has to change at each stage, makes the case that leadership fluency is not optional, and gives you the honest signs that you are stuck.
I’ve delivered enough of these sessions to recognise the moment. It’s about forty minutes in, the room has warmed up, someone from finance has just discovered they can get a variance summary in twelve seconds, and there’s a genuine buzz. Afterwards the sponsor says something like “that was brilliant, I think we’re really getting there.”
Six weeks later I check in and almost nothing has changed. Same processes, same documents, same Tuesday.
That gap has a name, and getting the name right matters, because awareness and fluency need completely different interventions. Awareness is knowing a tool exists and having tried it. Fluency is when the work itself has been rebuilt around the tool and using it stops being a decision.
Most companies have bought the first thing and are reporting it as the second.
You can see the gap clearly in the aggregate data. Stanford’s AI Index Report 2026 found organizational AI adoption rose to 88% of surveyed organizations, and generative AI is used in at least one business function at 70% of them. In the same breath it reports that “AI agent deployment was in the single digits across nearly all business functions,” and that across most functions “a majority of respondents reported no agent use at all”[3]. Nearly everyone has AI somewhere. Almost nobody has changed how work runs.
The US Census Bureau, which surveys actual businesses rather than executives at conferences, is even more sobering. Between December 2025 and May 2026, overall AI usage among US businesses hovered between 17% and 20%[4]. And when the Census asked workers directly, 55% said they’d used AI on the job for at least one task, but only 24% of those users had used it every day in the previous week, while 30% hadn’t used it at all that week[5].
That’s the whole argument of this article compressed into one finding. If skills were the constraint, more training would fix it. Something else is holding it.
Maturity models are usually consultant wallpaper. This one is only useful if you can honestly place yourself, so I’ve written each stage as a description of behaviour you can observe rather than a set of capabilities you can claim.
Be warned that most organizations rate themselves one stage higher than an outside observer would.
Future Factors’ maturity progression. This is an author’s framework, not survey data. Stage names differ across published models: Stanford’s AI Index uses “Not using / Experimenting / Piloting / Scaling / Fully scaled”[3].
Everyone’s heard of it. A handful of people use ChatGPT on their phone for personal things and occasionally for work, quietly, because they’re not sure if they’re allowed. There’s no policy or the policy is “be careful.” No process has been redesigned. Leadership talks about AI in all-hands meetings using the word “journey.”
The tell: if you asked ten people to name one task that is now done differently because of AI, you’d get ten blank looks and one person describing something they do personally.
Licences have been bought. There’s a pilot, or three. A few enthusiasts have built genuinely clever things and told almost nobody. Training has happened, usually once. Someone is tracking usage, and the number went up and then flattened.
This is where the largest number of organizations are stuck, and it can last for years because it looks like progress from the inside. There’s activity. There are dashboards. What there isn’t is a single process that has been rebuilt.
The tell: your AI story is a list of tools and pilots rather than a list of changed processes.
At least a few named, recurring workflows genuinely run differently now, and would visibly break if you took the tool away. There are standards for what “good” looks like when AI helped. Managers reference AI in normal work conversations rather than in AI conversations. New joiners are taught the AI-assisted version of the process as the process.
The tell: you can name the process, the person who owns it, and what it looked like before.
Here’s where it gets interesting, because fluency is not “uses AI for everything.” Microsoft’s 2026 research on its most advanced users found 86% of them say they treat AI output as a starting point rather than a final answer and that they “stay responsible for the thinking.” Those same users are more likely than others to say they intentionally do some work without AI to keep their skills sharp (43% versus 30%) and to pause before starting to decide what should go to AI versus a human (53% versus 33%)[1].
Fluency is judgement about allocation. A fluent organization is one where people are as good at deciding what not to hand over as they are at prompting.
The tell: people in your organization can articulate, without being asked, a category of work they deliberately keep away from AI, and give you a reason that isn’t “policy.”
This is Future Factors’ own framework, and it’s the thing I’d keep if I had to throw out everything else in this article:
Tool x Workflows x Behavior = AI-powered professional.
The multiplication signs are doing all the work. This isn’t a checklist where you collect points for each item. If any one of the three is zero, the product is zero, however strong the other two are.
Tool is the licence, the access, the tenant configuration. It’s the only part you can complete by spending money, which is exactly why it’s the part organizations complete first and fastest.
Workflows is whether a specific, named, recurring task that a specific role does has actually been redesigned. Not “you could use AI for that.” A real process, done differently, with the old version retired. In our experience auditing stalled programs, this is the missing variable roughly eight times out of ten.
Behavior is whether the new way survives a busy week. It’s the variable people most underestimate because it feels soft, and it’s measurable and stubborn.
Map that back onto the four stages and the model explains the sticking points precisely. Awareness is Tool only, so the product is zero. Experimentation is Tool plus scattered Behavior with no Workflows, so still zero. Integration is where Workflows finally becomes non-zero and the multiplication starts producing something. Fluency is when all three are high and reinforcing each other.
What gives me confidence this framing isn’t just a tidy metaphor is that Microsoft’s research arrives somewhere very similar from a completely different direction. Its 2026 Work Trend Index found that organizational factors like culture, manager support and talent practices account for more than twice the reported AI impact of individual factors like mindset and behavior, 67% versus 32%[1]. Microsoft notes these are statistical associations rather than causal effects, and that caveat is worth carrying. But “organizational” here is largely another word for Workflows.
The same report segments AI users in a way that maps almost directly onto the equation. Only 19% are “Frontier,” where organizational capability and individual readiness are both high. 16% are “stalled,” with low capability and limited organizational support. And 10% fall into what Microsoft calls “blocked agency,” where “individuals have built strong skills but lack the systems to apply them”[1].
Blocked agency is a high Behavior score multiplied by a Workflows score of zero. Skilled, motivated people producing nothing, because the organization around them never changed the work. If you have trained people well and seen no return, that’s very likely what you’re looking at, and more training will make it worse rather than better.
Each transition has a different bottleneck, which is why the same intervention works brilliantly at one stage and does nothing at the next. The most common mistake I see is an organization applying the stage-one remedy (awareness training) to a stage-two problem.
The bottleneck here is genuinely simple, and it’s mostly not a skills problem. People need to know they’re allowed, know what data they can and can’t put in, and have access to a tool that works with their actual files.
What moves it: a clear, short, written policy that says yes to specific things rather than a vague “be careful”; licences actually assigned; and one visible person doing it openly. What doesn’t move it: a two-hour introduction to large language models.
This is the hard one, and it’s where most organizations stall for years. The bottleneck is that nobody has authority to change a process.
What moves it: picking three named recurring tasks in one team, redesigning them properly with the people who do them, retiring the old version, and writing down what “good” looks like for output that AI helped produce. That last part is underrated. Without a quality standard, every AI-assisted deliverable becomes a debate.
What doesn’t move it: more pilots, more tools, another training round for the same people. Cisco’s 2025 AI Readiness Index found only one in three companies have a formal change management plan to guide employees through AI adoption, and only 32% have a process to measure the impact of their AI initiatives[8]. Those two gaps are the stage-two bottleneck restated in survey form.
The bottleneck at this stage is that the system still rewards the old behaviour.
Microsoft’s 2026 data is uncomfortably specific here. 65% of AI users fear falling behind if they don’t use AI to adapt quickly, yet 45% say it feels safer to focus on current goals than to redesign work with AI. And only 13% say they’re rewarded for reinventing work with AI when results don’t immediately follow. Microsoft calls this the Transformation Paradox: “Employees are ready to reinvent how they work, but the system around them, metrics, incentives, and norms, continues to reinforce the old way”[1].
What moves it: changing what gets recognised. Making it explicitly safe to try a redesign that doesn’t work. Teaching allocation judgement rather than more prompting. What doesn’t move it: a fifth tool.
Here’s the pattern I’ve watched play out more times than I can count. An organization decides its people need AI skills, funds a training programme for individual contributors and middle managers, and exempts the leadership team on the grounds that they’re busy and strategic.
Then it wonders why nothing scales.
The data on this is unusually consistent across independent sources. Gallup’s survey of 22,368 US employees found 69% of leaders use AI at least a few times a year, against 55% of managers and 40% of individual contributors. Frequent use among leaders has risen from 17% to 44% since mid-2023, while individual contributors moved from 9% to 23%[6]. So leaders are ahead on personal usage. That part is fine.
The problem is what happens between personal usage and organizational permission. Microsoft found that only 26% of AI users say their leadership is clearly and consistently aligned on AI[1]. Leaders themselves report a much rosier picture than their staff: 81% of leaders said they feel safe suggesting new ways of working with AI, against 67% of employees, and 78% said their managers create space for AI experimentation, against 59% of employees[1].
Leaders are using AI privately and reporting a culture their people don’t experience.
The mechanism isn’t mysterious, and it’s been measured. Microsoft cites a study of 1,800 workers globally which found that when managers actively modelled AI use, employees reported a 17-point lift in reported AI value, a 22-point lift in critical thinking about their AI use, and a 30-point lift in trust in agentic AI. Where managers created psychological safety around experimentation, employees reported up to 20 points higher AI readiness and value, and were 1.4 times more likely to be high-frequency users[1].
The comparison between Microsoft’s most advanced users and everyone else is the cleanest statement of it. Frontier Professionals are far more likely to say their manager openly uses AI (85% versus 64%), sets quality standards for AI work (83% versus 57%), creates space for experimentation (84% versus 61%), and encourages more ambitious work redesign (87% versus 61%)[1].
Look at that second item, because it’s the one nobody thinks of as leadership work. Setting quality standards for AI-assisted output. That’s a leader deciding what “good” means now, and it’s the single most useful thing a director can do for a team stuck at stage two.
Not much, honestly, and less than most leaders fear. Three things:
IBM’s 2025 CEO study found 64% of CEOs say the risk of falling behind drives them to invest in some technologies before they have a clear understanding of the value those solutions bring[10]. That’s an honest admission, and it explains a lot of stage-two spending. The way out isn’t a better tool selection process. It’s leaders who understand the work well enough to say which processes should change.
Stuck organizations rarely feel stuck. They feel busy. Here are the honest diagnostics, in rough order of how reliably they’ve predicted a stall in our work.
You can name tools but not processes. Ask your leadership team to describe your AI progress. If the answer is a list of products and pilots rather than a list of processes that run differently, you’re at stage two regardless of the spend.
Your usage graph spiked and flattened. Classic signature of training without workflow redesign. The spike is curiosity. The flatline is the absence of a reason to come back.
Nobody has retired an old process. This is the sharpest single question I know. Real integration means something got switched off. If both the old and new ways still exist, people under pressure will use the old one, every time.
Your AI wins are all individual. One person in ops built something great. Nobody else uses it, and if they left it would vanish. Individual brilliance without a system is a stage-two hallmark and it’s fragile.
You don’t have a quality standard. If your team can’t tell you what makes an AI-assisted document acceptable, every deliverable becomes a negotiation and people quietly avoid the friction by not using AI at all.
Nobody has been rewarded for a failed experiment. Not “we say failure is okay.” Has it actually happened, visibly, in the last six months? If not, your people have correctly worked out what’s safe.
Your training and your incentives disagree. You’ve trained people to redesign work, and you measure them on the throughput of the old process. They will optimise for the measurement. They always do.
Leadership exempted itself. Covered above, but worth repeating as a diagnostic. If your exec team was excluded from AI training because they’re busy, that decision is now your ceiling.
Work on the transition you’re actually facing rather than the one you’d like to be facing. Trying to build fluency while you’re at awareness produces a very expensive stage-two organization.
Write a one-page policy that says yes to specific named things and no to specific named things. Assign licences to a whole team rather than sprinkling them across volunteers. Get one senior person using it openly and talking about it in normal meetings. Run one short, practical session built on that team’s real documents, not a general introduction.
You’ll know you’ve moved when people are using it in the open and asking questions in a group channel rather than privately.
This is the transition worth putting real weight behind, because it’s where almost everyone stalls.
Choose one team. Pick three recurring tasks they genuinely do every week. Redesign each one properly with the people who do it, write the new version down, and retire the old one. Then write your quality standard: one page on what acceptable AI-assisted output looks like here.
Give it a named owner with protected time. Not a committee, since a committee has no calendar. Measure two things only: active users as a share of enabled users, and whether the three processes are still running the new way in ninety days.
Resist adding tools during this phase. The pull to solve a workflow problem by buying something is very strong and it’s almost always wrong.
You have proof it works. The job now is to make the new behaviour the default rather than the exception.
Put AI-assisted process redesign into how managers are assessed. Reward an attempt that failed, publicly, at least once. Teach allocation judgement: when to hand work over, when not to, how to check what came back. Build the AI-assisted version of a process into onboarding, so new joiners never learn the old one.
Then keep the material current, because a fluent organization running on a two-year-old curriculum degrades quietly. That maintenance discipline is a whole subject in itself, and it’s the focus of the companion piece to this article on building an AI upskilling program that doesn’t go stale.
Awareness to experimentation can happen in weeks. Experimentation to integration takes two to three quarters if someone owns it, and forever if nobody does. Integration to fluency is measured in years and is mostly about incentives rather than capability.
Anyone promising you organizational AI fluency in a quarter is selling you a stage-one intervention with a stage-four name on the box.
If you want to go deeper on the underlying capability question, our plain-English guide to what AI literacy actually means is a good place to start, and the step-by-step playbook for training your team on AI covers the mechanics of running the sessions. For the leadership gap specifically, our piece on how far apart leaders and employees are on AI usage puts numbers on it.
Awareness means people know AI exists and some have tried it, while the work itself is unchanged. Fluency means named processes have been redesigned around AI, reaching for it is automatic rather than a decision, and people exercise judgement about what to keep away from AI. The gap shows up in aggregate data: Stanford’s AI Index found 88% of surveyed organizations use AI while agent deployment stays in single digits across nearly every business function[3], and IBM found CEOs reporting only 25% of their workforce uses AI regularly despite 86% believing their people have the skills[7].
Four: Awareness (people know it exists, use is hidden and individual, no process has changed), Experimentation (licences and pilots, visible enthusiasts, still no redesigned workflow), Integration (named recurring workflows genuinely run differently, quality standards exist, the old version has been retired), and Fluency (AI use is automatic, and so is deciding when not to use it). Most organizations stall in Experimentation for years because it looks like progress from the inside. Published models use different labels: Stanford’s AI Index charts deployment as Not using, Experimenting, Piloting, Scaling and Fully scaled[3].
It states that Tool x Workflows x Behavior = AI-powered professional. The multiplication is the point: if any variable is zero, the result is zero regardless of the other two. Tool is the licence and access, which is the only part you can buy. Workflows is whether a specific recurring task has actually been redesigned. Behavior is whether the new way survives a busy week. Awareness is Tool alone. Experimentation is Tool plus scattered Behavior with no Workflows. Both multiply out to zero, which is why organizations at those stages see activity but no return.
Both, and leadership is the constraint more often. Gallup found 69% of leaders use AI at least occasionally against 40% of individual contributors[6], so personal usage is not the gap. The gap is organizational permission: only 26% of AI users say their leadership is clearly and consistently aligned on AI, and leaders report far more psychological safety around AI than their employees experience[1]. Microsoft’s most advanced users are much more likely to say their manager openly uses AI, sets quality standards for AI work, and creates space to experiment[1].
Ask three people in different teams to name one thing they do differently because of AI, and what it replaced. Specific answers with named replacements mean you have reached integration. Enthusiasm without specifics means experimentation. Personal-life examples mean awareness. Other reliable signals: your AI story is a list of tools rather than processes, your usage graph spiked and flattened, no old process has ever been switched off, there is no written quality standard for AI-assisted work, and nobody has been visibly rewarded for an experiment that failed.
This guide combines Future Factors’ work moving corporate teams through AI adoption with primary research from Microsoft’s Work Trend Index, Stanford HAI’s AI Index, the US Census Bureau, Gallup, IBM, Deloitte and Cisco. Every figure was traced to the publishing organization’s own report rather than to secondary coverage. Widely-circulated statistics that could not be traced to a primary source, including the frequently-quoted “95% of AI pilots fail” line, were deliberately excluded: that figure is a press paraphrase of a preliminary working paper whose own wording is materially different.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.