You’ve been asked to lead AI adoption in your company. The strategy is in place, licences are secured, training has been launched and employees have heard why AI matters. At first, things look promising. Usage starts to climb, people begin experimenting and the early adopters create a sense of momentum.
Then, a few months later, the curve flattens.
Some employees are using AI every day, while others barely touch it. Leadership starts asking why adoption is not moving faster and what you are going to do about it. Suddenly, the exciting AI initiative starts to feel a little more like a hot potato.
If this sounds familiar, you are far from alone.
Organisations are investing heavily in AI, but turning that investment into sustained use and business value is proving harder than simply deploying the technology. BCG reported in 2026 that only a small share of companies in its research were generating substantial value from AI, while a much larger group reported little or no material value. Its broader point was that successful AI transformation depends on what happens around the technology as much as on the technology itself: how people work, how processes are organised and whether new behaviours become established.[1]
The gap between simply having AI and actually using it is visible in more controlled research as well. A six-month study followed more than 7,000 knowledge workers across 66 companies who were given access to Microsoft 365 Copilot. Just having access did not mean people used it regularly. The typical employee used Copilot in only about 4 out of 10 weeks. In the final three months of the study, 1 in 5 did not use it at all, while only 1 in 20 used it every week. [2]
That is the important point: providing access is not the same as creating adoption.
Imagine three employees who have access to exactly the same AI tool and have attended exactly the same training. One uses it every morning to prepare for meetings, analyse information and draft documents. Another is interested in using it but remains unsure about what company information can safely be entered, so whenever a real task comes up, they decide it is safer not to use AI. The third knows perfectly well how the tool works, but when a deadline is approaching, automatically opens the same spreadsheet or template they have used for the past five years.
The technology is the same. The training may have been the same. The behaviour is not.
From a usage dashboard, the second and third employees might simply appear as “low users”. But what is behind their behaviour is completely different. One may be dealing with uncertainty. The other may be dealing with habit. Someone else might lack confidence, not see enough value for their role or simply find AI too cumbersome to use alongside everything else they have to do.
Gallup’s 2026 research reinforces this point. Among employees working in organisations where AI was already available, people who rarely or never used it gave a range of different reasons. Some preferred to continue working in familiar ways. Others questioned whether AI was useful for their particular job, while concerns about ethics, privacy, security and compliance also featured prominently.[3]
So low adoption is not one problem. It is an outcome that can be produced by many different problems.
And that is where behaviour gets interesting.
AI adoption depends on both the person and the environment
When adoption is lower than expected, it is tempting to search for one explanation. Perhaps employees need more skills. Perhaps managers are not encouraging AI enough. Perhaps the tools are poorly integrated. Perhaps employees simply do not trust AI.
Any of those explanations could be right. But behaviour rarely has one neat cause.
People bring their own knowledge, confidence, beliefs, habits, experiences and concerns to AI. At the same time, they operate inside an organisation that shapes what is easy, difficult, encouraged, permitted and normal.
Both sides matter.
Consider someone who has excellent access to AI, clear policies and a supportive manager, but does not see how the technology adds value to their particular role. They tried it several times, received mediocre results and gradually returned to their old way of working. Better workflow integration alone may not solve that problem. The barrier may have more to do with perceived usefulness, confidence or previous experience.
Now imagine the opposite employee. They are highly motivated, know how to use AI well and can immediately see where it would help. But using it means moving information between several systems, policies are difficult to interpret and their manager rarely gives the team time to experiment. Their willingness is not the problem. The environment is making the behaviour unnecessarily difficult.
And then there are employees somewhere in between: they have some skills but little confidence, supportive managers but unclear rules, good tools but no established routine for using them.
Same low usage. Different behavioural conditions.
Gallup’s data show how strongly some of those organisational conditions are associated with actual use. Among employees who strongly agreed that AI fitted well with the systems and processes they already used, 88% were frequent AI users, compared with 55% of those who did not strongly agree. Frequent use was also substantially higher among employees who felt their manager actively supported AI, who felt encouraged to experiment and who considered their organisation’s AI policies clear.[3]
Those findings do not prove that changing one policy or asking a manager to be more supportive will automatically increase adoption. But they are a useful reminder that we should not treat AI use as something determined by the employee alone.
McKinsey’s 2026 research points to the same tension from another direction. In its survey of 750 employees and leaders already using AI, 70% of respondents said they felt personally prepared to work with AI. Yet only 27% of leaders believed their organisations were ready for the broader shifts required. McKinsey’s analysis also found organisational readiness - including workflows, leadership practices, operating models and cultural norms - was more strongly associated with reported enterprise value than personal readiness alone.[4]
In other words, employees can be ready before the organisation around them is.
This is where the familiar saying culture eats strategy for breakfast becomes relevant. But culture should not be treated as an abstract explanation for everything that goes wrong. It is partly created through very practical experiences: what people have time to do, what managers encourage, what gets rewarded, what policies permit, what processes make easy and what colleagues around them are doing.
Tell employees that experimenting with AI is a strategic priority, but give them no time to experiment, and the message loses some of its power. Tell them to incorporate AI into everyday work, but place the tool outside their normal workflow, and familiar routines will often win. Encourage experimentation while leaving employees unsure about what information can safely be entered, and avoiding AI may be the perfectly reasonable choice.
AI adoption is therefore not only a technology rollout; it is an organisational change challenge. Employees may need to change how they work, build new skills, develop trust in the tools and rethink parts of their professional role. At the same time, the organisation around them may also need to change - from workflows, incentives and expectations to how time is allocated, how experimentation is supported and what behaviours are recognised as valuable.
At AI Modus, we find it useful to distinguish between individual barriers and organisational barriers when looking at AI adoption.
Individual barriers can include not knowing where AI is useful, lacking confidence or practical skills, finding it difficult to evaluate outputs, distrusting the technology, not seeing enough value in a particular role, forgetting to use AI or simply falling back into established habits.
Organisational barriers can include poor access, tools that do not fit existing systems, unclear policies, insufficient time to experiment, competing priorities, weak manager support, unclear strategic direction or few opportunities to learn from colleagues who are already using AI successfully.
Neither category exists independently of the other. Organisational conditions can influence individual behaviour, and individual experiences can change how people respond to those conditions.
Someone may begin highly motivated but repeatedly encounter unclear policies and cumbersome processes. After several frustrating attempts, their motivation falls. Someone else may work in an organisation with excellent tools and strong leadership support but still not understand how AI applies to their specific job. Another employee may have the skills, access and motivation, but under pressure continue to choose the old method because it has become automatic.
So the useful question is not simply: “Why aren’t our employees using AI?” It is:“What is making the behaviour easier or harder?” Usage is a measure. It is not a diagnosis.
Suppose the next leadership report shows that only 30% of employees regularly use your organisation’s AI tools. That number matters. It tells you something about what is happening. But what should you do with it?
Should you run more training? Develop role-specific use cases? Rewrite the AI policy? Ask managers to encourage experimentation? Give employees more time to practise? Integrate AI more deeply into existing systems?
The usage number cannot tell you.
The same applies when looking at differences between teams. Imagine that Team A uses AI considerably more than Team B. It is easy to conclude that Team A is simply more innovative, motivated or open to change.
But perhaps Team A happens to perform tasks for which current AI tools are particularly useful. Perhaps its manager regularly demonstrates how they use AI themselves. Perhaps employees have clearer guidance, more time to experiment or colleagues they can ask for help. Perhaps Team B has received exactly the same licence and training but works under completely different conditions.
Those differences are not noise around the adoption problem. They may be the adoption problem.
So, when AI adoption stalls, the next step should not automatically be another intervention. It should be to look beneath the adoption number.
A useful starting point is to investigate three broad questions.
First, can people use AI effectively? Do employees understand where it is useful in their work? Do they have the knowledge and confidence to use it? Can they evaluate outputs and recognise when human judgement is still needed? If the answer is no, training and practical experience may be exactly what is required.
Second, does the environment support AI use? Are the right tools available? Are policies clear? Does AI fit naturally into existing workflows, or does using it create extra work? Is there time to experiment? Do managers support it? Can employees learn easily from colleagues who already use AI well?
Third, do people see enough reason to use it - and has it become part of their normal behaviour? Do employees see concrete value for their own role? Do they trust AI enough to use it? Does it fit with how they think about doing their job well? When work becomes busy, does AI come to mind, or does the familiar way of working automatically take over?
These questions give leaders somewhere more useful to start than simply asking why “people are not adopting”. The next step is to gather evidence.
Talk to employees rather than only talking about them. Look at usage patterns alongside employee survey data. Compare teams where adoption differs. Use interviews or focus groups to understand what lies beneath the numbers. Examine what someone actually has to do when they try to use AI during a normal working day.
And distinguish evidence from assumptions.
“Our people need more training” is not yet a diagnosis. It is a hypothesis.
“Managers need to encourage AI more” is a hypothesis.
“Employees just need to understand the benefits” is a hypothesis.
The point is not to spend months diagnosing every possible influence before doing anything. It is to gather enough information to make a better decision about where to look first.
This is also the logic we have built into AI Modus: the platform gives organisations a structured way to gather behavioural data about AI adoption, distinguish between individual and organisational barriers and see where patterns differ across groups or teams. It is one way of making this type of diagnosis more systematic, particularly when the organisation is too large to rely on individual conversations alone.
But the principle applies regardless of the method used: diagnose before you intervene.
Training is a good solution when capability is the barrier. If people know how to use AI but remain uncertain about what is permitted, clearer policies may matter more. If employees see the value but AI sits awkwardly outside their workflow, integration may be the priority. If useful practices already exist but remain isolated within a handful of teams, making those practices visible and creating opportunities for peer learning may help.
Different barriers require different responses.
There is another reason this matters. The goal is ultimately not simply to increase log-ins. McKinsey’s research suggests that higher individual use alone does not necessarily translate into wider organisational value. Organisations also need to rethink workflows and ways of working around what AI makes possible.[4]
So when the adoption curve flattens and someone asks what you are going to do next, perhaps the best answer is not another initiative just yet.
Start with the question underneath the number: What is shaping this behaviour?
Because usage tells you what is happening. Understanding the barriers helps you decide what to change.
References
Written by