88% of organisations now use AI in at least one business function. Only 39% report any measurable EBIT impact. That is the most expensive gap in business right now, and it is where AI adoption stalls. Notice that it’s not in the pilot phase, not at the infrastructure level, but somewhere between deployment and results. 🤔
A research paper published in February 2026 by Harvard Business Review offers one explanation for the gap. Fractional Insights and Ferrazzi Greenlight surveyed more than 3,000 employees across the US and Europe and found that employee anxiety about AI (what they call “AI angst”) is both widespread and actively driving the wrong kind of behaviour.
About 8 in 10 employees reported strong concern around at least one AI-related fear. 65% worry about being replaced by someone who uses AI better than they do. 61% worry that AI makes others think they don’t bring unique value anymore.
The research is credible, the data is solid, and the psychological framing is genuinely useful. But after working with organisations across Europe, the US, Asia and the Middle East, on AI automation implementation, I think the diagnosis is missing a layer. Fear is not the root cause. It is what shows up when leadership hasn’t done its job properly.
Learn Why AI Adoption Stalls
The compliance trap
The most counterintuitive finding in the HBR research is this: employees with high AI anxiety reported using AI for an average of 65% of their work, compared to 42% for those with low anxiety. The anxious ones use AI more. But they also show more than twice the resistance score. 🤷♂️
What you have, in practical terms, is a workforce going through the motions. Licences activated, tools used, dashboards showing healthy engagement numbers, and almost no real value being generated. The HBR researchers call it “performative rather than participatory” usage. I call it compliance theatre. 😉
And it is precisely why usage metrics are a completely useless proxy for AI adoption success.
I have seen this pattern repeatedly.
A company runs an AI rollout, announces it internally with a fair amount of fanfare, tracks adoption through logins and feature usage, and then, at the next board review, wonders why productivity hasn’t moved. The tools are being used. Just not in any way that changes how work actually gets done.
The reason this happens is straightforward. When you push AI on people without giving them a clear process context, a genuine training programme, or any reassurance about what their role will look like after implementation, they learn to use AI in ways that seem acceptable without threatening their position.
They summarise documents.
They clean up emails.
They generate first drafts that they would have written anyway.
Safe, low-risk, low-impact usage that ticks the adoption box without changing anything.
Where the fear actually comes from
The HBR research categorises employees into four groups based on their AI beliefs and anxiety levels: Visionaries, Disruptors, Endangered employees, and Complacent ones. It is a useful framework for HR teams thinking about change management.
For C-suite decision-makers with an AI budget, though, the more important question is: why are so many people in the Disruptor and Endangered categories to begin with?
The answer is rarely individual psychology. It is organisational design.
When a company launches an AI initiative without redesigning the workflows around it, employees have no choice but to feel threatened. The tool arrives, but the job description doesn’t change.
Nobody explains what success looks like for a human who now works alongside AI.
No one addresses what happens to the people whose tasks get automated.
The ambiguity itself creates the anxiety.
McKinsey’s 2025 State of AI report identifies workflow redesign as the single strongest factor correlating with actual EBIT impact. Stronger than any technical capability, tool selection, or budget size.
The 6% of organisations classed as “high performers” are more than twice as likely to have fundamentally redesigned their workflows around AI. They didn’t just add AI to existing processes but rebuilt them.
Most organisations do the opposite. They treat AI as an add-on layer, keep the old workflows intact, stick AI tools on top, and then measure usage to prove adoption is working. The fear employees feel in that situation is entirely rational. They are watching a technology enter their workplace, and nobody can tell them what their job actually looks like now.
Why big AI transformations fail
The HBR paper points out that organisations in technology and financial services show both the highest belief in AI’s value and the highest AI angst. These are sophisticated people who understand the technology well enough to see exactly what it threatens. That combination, high awareness plus high personal risk perception, produces the fear-driven usage pattern the research describes.
From where I sit, this is also the population most likely to be working inside organisations that have launched ambitious, enterprise-wide AI transformation programmes. And those programmes (the big-budget, all-at-once, multiple-function rollouts) are among the most reliable ways to generate AI angst at scale.
A large transformation project creates a long window of uncertainty. People spend months, sometimes years, unsure of what the end state of their role looks like. Decisions get made at the top and filter down with minimal context. Training, if it happens at all, is generic. It covers the tool, not the work. And because the project is large and visible, the stakes feel enormous for everyone involved.
The organisations I have seen actually capture AI value don’t operate this way. They run small, focused implementations against specific process problems. One workflow, one team, one clear outcome.
They measure that, learn, adjust, and only then do they move to the next one. The uncertainty window is short, results are visible quickly, and the people involved understand exactly what changed and why, because the scope was narrow enough to explain properly. ✌️
Small bets beat large transformation projects. If you think it’s because the ambition is lower, you would be wrong, because the risk is contained and the feedback loops are faster.
The expertise problem that nobody wants to fund
There is a third factor that the HBR research doesn’t address directly, but that shows up everywhere in practice: the gap between having AI tools and having the expertise to deploy them well.
BCG’s research puts 60% of companies in the category of generating no material value from AI despite real investment. The McKinsey data tells the same story from a different angle. The common thread is not a lack of tools, but a lack of people who understand both the technology and the domain it is being applied to well enough to make the combination work.
AI is extraordinarily good at explicit, codified knowledge. It is genuinely poor at tacit expertise: the judgement, the contextual pattern recognition, the instinctive prioritisation that experienced professionals carry in their heads and rarely write down anywhere.
When organisations deploy AI without that domain expertise in the loop, they get competent-looking outputs that miss the point. The tool produces something that looks right but isn’t, and the team doesn’t always know how to identify the difference.
This is why the T-shaped professional (deep domain expertise combined with genuine AI literacy) has such a structural advantage right now. AI amplifies what that person knows. For everyone else, it produces an impressive quantity of mediocre output.
The implication for leadership is that AI training cannot be a one-day workshop on prompt engineering.
It needs to connect the tool to the specific work context of the people using it. That requires investment in time, not just budget. And it requires accepting that training must come after the process has been manually mapped and understood, not before.
What C-suite should actually do with this
The HBR research recommends three leadership shifts: recognise industry-shaped risk, stop treating usage as a proxy for buy-in, and design for learning before scale. All three are right. Here is how they translate into practice.
Map and optimise the process manually before you automate anything. If you can’t explain a workflow clearly without AI, you cannot automate it well with AI. The manual mapping step also forces a conversation about what the work actually is, which is the same conversation that reduces anxiety, because people can see what changes and what stays the same.
Define the human role explicitly in every AI implementation. The fear the HBR research documents is, at its core, a question of “where do I fit?”. Answer that question for every team before deployment, not after.
The AI high performers in McKinsey’s data are significantly more likely to have defined human-in-the-loop validation processes (65% vs 23% of everyone else). That is not a coincidence. It is the mechanism through which people stay engaged rather than compliant.
Start with one thing and prove it. Pick the highest-friction, lowest-risk workflow in a single team.
Automate it properly, and measure the outcome. Make the result visible to the wider organisation.
That single visible success does more for genuine adoption than any amount of top-down mandate, because it makes the benefit concrete and the threat abstract. Employees can see that automation improved the work, not eliminated the worker.
Invest in real training, not tool familiarisation. Generic AI training produces generic AI usage. Role-specific, workflow-specific training that ties the tool to actual decisions and actual outputs is the difference between compliance and genuine capability.
The HBR research is a useful mirror for organisations that have been measuring adoption and not understanding their results. The anxiety is real, and it matters. But anxiety is a symptom of something leaders created, and it is within leadership’s power to address it… not through psychographic profiling, but through better process design, honest communication, and a willingness to think small before thinking at scale.
The 6% of organisations actually winning with AI right now didn’t manage their employees’ feelings about AI particularly well. They built processes that gave people something concrete to work with, a clear role in the outcome, and visible proof that the investment was worth making.
That is where the gap closes.
Leave a Reply