Guessing at your next AI adoption intervention is expensive
Picture a 400-person operations group six months into an AI pilot. Usage reports show plenty of logins, but the VP of transformation still can't tell the board whether the AI adoption strategy behind the rollout is actually working.
The usual response is to roll out more training or simply mandate use. Both are reasonable starting points: training does close real skill gaps, and a clear policy removes ambiguity about whether AI use is expected at all.
But neither move answers the question that actually decides whether it works: why are these specific people not using the tool? Pick the wrong lever and you spend a quarter's budget on a problem your workforce doesn't have. An AI adoption strategy has to start by finding that out, not after the spending.
We're skeptical of any rollout plan that opens with a training budget before anyone has asked employees what's actually stopping them. Training closes a skills gap; it does nothing for a workflow that doesn't fit, or a manager who never mentions the tool.
That support gap is wider than most leaders assume. Just over a third of U.S. workers said in late 2025 they had the training and resources needed to use AI at work, according to Jobs for the Future's national worker survey, weighted to match the U.S. workforce, down from the year before.
“the new survey shows growing concern about the pace of AI adoption and a clear gap in employer support.”
— Ben Pring, Vice President of JFF's Center for Artificial Intelligence & the Future of Work · Source: Worker AI Anxiety Grows as Employers Lag: JFF Survey
Sources: Worker AI Anxiety Grows as Employers Lag: JFF Survey
What employee surveys reveal about AI usage and barriers
Start with scale. Gallup's February 2026 study of 23,717 U.S. employees at organizations that make AI tools available found that managerial support of AI use, and how well a tool fits someone's existing workflow, predicted adoption more strongly than a person's role did.
That's a useful place to start an AI adoption roadmap: not with usage counts, but with the levers survey respondents themselves point to. The rest of this section breaks down who's actually using AI, and what's stopping the people who aren't.
Sources: AI in the Workplace: What Separates Adopters and Holdouts
Who uses AI often, and why
Self-reported usage numbers often undercount what's really happening. A 2026 analysis summarizing an earlier 2022 MIT Sloan Management Review and Boston Consulting Group survey of 1,741 managers found that about two-thirds of respondents said they used AI minimally or not at all, until the question offered concrete examples of AI-assisted tasks, after which many realized they already qualified.
The pattern repeats elsewhere. In the third wave of the Konstanz AI study, which surveyed 1,105 employees in May 2026, most AI use turned out to be employee-initiated rather than something the organization formally introduced, a gap Florian Kunze, who heads the study, described this way:
“In many organizations, the major AI revolution has not unfolded as a planned transformation process so far,”
— Florian Kunze, Head of the Konstanz AI study and professor of organizational behavior at the University of Konstanz · Source: Workplace AI adoption remains slow, informal and uneven, survey finds
In our view, that gap between official rollout and real use is where a leader should look first: employees already improvising with AI are telling you which tasks matter, often more reliably than any pilot plan did.
Sources: Report finds employees embrace AI when they see its value | MIT Sloan, Workplace AI adoption remains slow, informal and uneven, survey finds
What separates non-users from occasional users
Not every non-user is the same kind of non-user. Gallup's 2026 survey found that employees who never use AI most often cite ethical opposition as their reason, a different objection from the practical friction infrequent users describe.
That distinction matters for what you do next. Treating ethical objectors and time-pressed occasional users as the same audience wastes a training budget on people who object to AI on principle, not on skill.
Questions worth asking when you segment a result like this, offered here as suggestions rather than a validated instrument:
- Does this person object to AI in principle, or to how it was introduced?
- Is the barrier about skill, about time, or about trust in the output?
- Would a visible use case from their own team change the answer?
AI Adoption Score's AI Adoption Barriers Survey is built to separate those reasons instead of lumping every non-user into one bucket.
Sources: AI in the Workplace: What Separates Adopters and Holdouts
Turning perception data into an intervention decision
Surveys also ask employees whether AI changed their work for the better. Treat those answers as a signal, not a verdict: people are reporting a perception, not a measured productivity gain, and nothing here tracks that causal link directly.
The signal still matters. In a 2022 global survey of 1,741 managers by MIT Sloan Management Review and Boston Consulting Group, organizations were 5.9 times as likely to report significant financial benefit from AI when their employees said they personally got value from it.
Ask employees directly which tasks they find AI useful for, and use the answer to prioritize where to expand access. If nobody names a task AI helps with, don't expect a mandate alone to change that.
Use the strongest answers to find candidate workflows worth documenting and sharing with other teams, rather than reporting them upward as a win on their own.
AI Adoption Score's Trust in AI Survey measures a related signal: how much a team believes in AI as a long-term force, including trust in current tools and confidence in an AI-shaped future, which can sit alongside usage and skills data when you're deciding where to look next.
Sources: Achieving Individual — and Organizational — Value With AI
Mandates, trust and manager support as competing levers
Mandates work, at least on paper. In MIT Sloan Management Review and Boston Consulting Group's 2022 survey of 1,741 managers, requiring AI use was linked to roughly triple the likelihood of regular use compared with trust-building efforts alone, though that figure is now several years old and the finding comes from a manager-only sample, not the broader workforce.
A mandate without context breeds quiet resistance, not real adoption. Pair it with what employees say they're missing: visible use cases from people like them, and a manager who talks about AI as part of the job, not an extra.
Common advice treats the mandate as the whole intervention. We don't: a mandate without addressed ethics concerns and manager backing just moves the resistance underground, where no login count will ever catch it.
This is also where manager support earns its place as a lever in its own right, not just a nice-to-have alongside training.
Sources: Achieving Individual — and Organizational — Value With AI
A practical framework for matching evidence to action
Once you've segmented the barrier, the choice of intervention gets a lot less subjective. The table below lines up an evidence pattern against the lever it points to, drawn from the surveys discussed above.
| Evidence pattern | What it usually means | Intervention to try first |
|---|---|---|
| Never-users cite ethical opposition | Objection to AI in principle, not a skill gap | Address the ethics concern with real use cases before any training spend |
| Infrequent users cite time or workflow friction | The tool doesn't fit how the task actually gets done | Redesign the workflow step, not the person |
| Low reported manager support | Managers aren't modeling or discussing AI use | Run a manager-enablement push before a broader mandate |
| Mandate exists but usage stays flat | Policy without trust or use cases behind it | Pair the mandate with visible use cases and manager backing |
| Employees report AI helped them personally | A candidate workflow worth a closer look | Interview a sample before scaling the use case |
We'd skip a company-wide mandate as the first move almost every time. Start with the segment reporting the clearest, most specific barrier, fix that, and only mandate once the fix is visible.
AI Adoption Score maps every person, team and department to one of five maturity levels across skills, usage, culture, impact, vision and governance, which gives each row above a place to land before you write the next line of your enterprise AI adoption strategy.
Sources: AI in the Workplace: What Separates Adopters and Holdouts, Achieving Individual — and Organizational — Value With AI
Pitfalls that derail evidence-based AI adoption strategy
A few mistakes show up again and again once leaders finally have survey data in hand.
That last point deserves a second look. An AI adoption strategy isn't a single decision, it's a loop: measure, pick a lever, act, then measure the same barrier again to see if it moved.
You don't need five interventions at once, and you don't need a bigger pilot. You need one segmented answer about who isn't using AI and why, and the discipline to check, afterward, whether your fix actually changed that answer. Start there.
Frequently asked questions
Who should own the decision about which AI adoption intervention to run next?
Whoever owns the rollout budget should own it, informed by the survey segmentation, not by whichever department asks loudest. In practice that's usually the digital transformation leader or Head of AI, working with the managers of the segments the survey flags.
How often should we re-run the employee survey after making a change?
A reasonable cadence is once per intervention cycle, roughly every quarter, rather than continuously. The point is to check whether the specific barrier you targeted moved, not to generate a new dashboard every week.
What if different teams show completely different barriers?
That's common, and it's exactly why segmenting by role, function and tenure matters before you pick one intervention. Run separate, smaller fixes per segment rather than one company-wide program built around the loudest group's complaint.
Is a single survey enough to prove an intervention worked?
No. Treat one result as a baseline, then re-run the same questions after your change to see whether the targeted barrier actually shifted. A single wave only tells you where you stand today, not whether an intervention caused a change. Pair that comparison with whatever usage or outcome data your organization already tracks.
Should we address ethics objections before or after a skills-training push?
Before, for the segment that cites ethical opposition specifically. Training spent on people who object in principle rarely changes their answer, so address the objection first, then revisit whether training is still needed.
Sources: AI in the Workplace: What Separates Adopters and Holdouts
