This is becoming a familiar challenge in many organizations. A small group of employees experiments enthusiastically and quickly finds ways to integrate AI into their work. Others use it occasionally. Some barely use it at all.
The natural response is often to do more. More training. More communication. More use cases. More tools. But what if those aren’t the things holding people back?
One of the biggest mistakes we can make when thinking about AI adoption is to treat it primarily as a technology rollout. Because while the technology is new, the actual challenge is very human:
We are asking people to change their behavior.
Research on AI adoption increasingly makes the same point. Organizations often approach AI as an engineering challenge and assume that once a useful system is available, people will eventually start using it. But successful adoption depends on how people integrate AI into their existing workflows, habits and ways of making decisions.
And this is exactly where behavioral science becomes useful.
The rational AI user and the real AI user
Imagine the perfectly rational employee. The rational employee learns that an AI tool can save 30 minutes on a task and immediately starts using it.
The real employee thinks: “Interesting. I’ll try it when I have a bit more time.” Then the deadline arrives, they open the same template they have been using for the last five years, and the new AI tool disappears from their attention.
The rational employee reads the company’s AI guidelines, understands what is allowed and starts experimenting. The real employee is not completely sure whether a particular piece of information can be entered into the tool, so decides it is safer not to use it.
The rational employee hears that AI will take over repetitive work and sees an obvious benefit. The real employee might also wonder: “If AI can do this part of my job, what does that mean for my expertise? My role? My value?”
Research on organizational AI adoption points to anxiety, distrust, perceived loss of control and concerns about job security and purpose as some of the psychological responses that can shape whether employees embrace or resist AI.
The point is not that employees are irrational or resistant to progress. The point is that people do not make decisions about AI by objectively calculating all the potential benefits and then selecting the most efficient course of action.
We bring habits, emotions, uncertainty, previous experiences, social influences and limited attention into the decision. In other words, we behave like humans.
This is what behavioral science is interested in
Behavioral science tries to understand why people behave the way they do. What motivates us to act and what stops us. How our environment shapes our choices. How the behavior of other people influences us. How habits, emotions and cognitive biases affect the decisions we make.
And importantly, behavioral science does not start from the assumption that people are perfectly rational decision-makers.
Herbert Simon described us as boundedly rational: our decisions are constrained by limited time, limited information and limited cognitive capacity. We do not carefully optimize every decision we make. Much of the time, we find something that works well enough and move on.
That matters enormously for AI adoption. Because from the organization’s perspective, the new AI-enabled workflow may clearly be better. But from the employee’s perspective, the old workflow already works. They know where to click. They know how long it will take. They know what the output will look like. They know what can go wrong.
The new way may be objectively faster once mastered, but right now it requires attention, experimentation and uncertainty. And on a busy Tuesday afternoon, the old way often wins.
AI adoption is a behavior change process
When organizations talk about “AI adoption,” the phrase can sound abstract. But what does adoption actually look like? It means someone who used to start a report from a blank page now asks AI for a first draft. Someone who spent 45 minutes summarizing meeting notes now uses AI to create the initial summary. Someone who previously searched through several documents manually now asks an AI assistant to help locate relevant information. Someone who has always solved a particular problem in one way starts experimenting with another.
In each case, a behavior has changed.
That is why buying licenses is not adoption. Giving employees access is not adoption. Completing an AI training course is not adoption. Those things can enable adoption. But adoption happens when people actually begin doing something differently in their everyday work.
And once we look at it this way, the problem changes.
Suppose an organization wants employees to use an approved AI assistant when preparing the first draft of an internal document. They have access to the tool. They have attended training. They know management supports its use. But they still rarely use it. Why?
Perhaps they do not know how to write a useful prompt. Perhaps they know how, but do not feel confident enough. Perhaps they simply forget about the tool when they are busy. Perhaps opening it requires leaving the system where they already work. Perhaps the existing workflow feels easier. Perhaps they are uncertain about what information they are allowed to enter. Perhaps they tried AI once, received a poor result and decided it was not useful. Perhaps their manager never uses it. Perhaps everyone around them still produces documents the old way. Perhaps using AI makes them feel that they are somehow doing less “real” work.
From the outside, all of these situations look the same: low AI adoption. Behaviorally, they are completely different problems.
More training is useful, when training is the problem
Organizations often respond to low adoption by providing more information. If people are not using the tool, explain its benefits again. If adoption remains low, organize another training session. If employees still hesitate, produce another guide, FAQ or list of use cases.
Sometimes this is exactly the right response. If people genuinely do not know how to use AI, training can remove an important barrier. But what if they already know how? Training will not necessarily solve a trust problem. A communication campaign will not necessarily solve a workflow problem. Another presentation about productivity will not necessarily solve uncertainty about what is permitted. And explaining how much time AI can save may do very little for someone who feels that using it threatens an important part of their professional identity.
Research on AI adoption makes a similar point: organizations often emphasize the potential benefits of AI such as efficiency and productivity while paying less attention to whether employees actually feel able, motivated and empowered to integrate AI into their work.
This is why behavioral diagnosis should come before intervention. Before deciding that people need more training, more communication or another tool, we should understand what is preventing the behavior from happening in the first place.
So what does it mean to wear behavioral glasses?
It means becoming more curious about what people actually do rather than what we think they should do.
If we want more employees to use AI, more often, in their everyday work, we need to ask: What is stopping them from using it today? Is it a lack of knowledge? A lack of confidence? Habit? Trust? Uncertainty? Effort? Social norms? And what would make the new behavior easier, more natural or more worthwhile?
Only then can we decide whether the answer is more training, clearer guidance, better integration into existing workflows, visible role modelling from managers, opportunities to experiment or something else entirely.
Behavioral science is not a magic wand for AI adoption. But it gives us a better way to understand what has to change for adoption to happen.
The technology can be ready before the organization is. AI tools are evolving extraordinarily quickly. Human habits usually do not. And that gap matters.
Organizations can deploy a new AI tool overnight. But changing routines, building confidence, developing trust and creating new ways of working takes more than turning on a license.
If we treat AI adoption only as a technology change, we risk repeatedly reaching for technical solutions to behavioral problems. More tools. More features. More training. More communication. Sometimes those will be exactly what is needed. But sometimes they will not.
If we want people to use AI, we first need to understand why they are not using it today. And that is why leaders navigating AI adoption may need behavioral glasses more than ever.
References
Written by