Why Does AI Adoption Stall, and What Does Behavioral Science Say About Fixing It?

Enterprise AI spending keeps climbing while returns lag behind it. In the one leading study of 15,000 employees across 29 countries, 88% of employees report using AI at work, but only 5% use it in ways that transform how their work gets done. The other 83% are largely doing what early automobile designers did when they built cars shaped like carriages: applying a new tool to old tasks in the old way.

Such anachronistic approaches produce healthy usage numbers and no change in output, but deployment metrics like license counts and login frequency cannot tell the two groups apart. Access to a tool and a changed way of working are different outcomes, and most rollouts were built to produce the first one. Closing that gap requires treating AI adoption as a behavior change challenge rather than a technology deployment. This is the principle behind Perceptyx’s Activating AI Adoption workshop.

Why Do AI Rollouts Stall After Deployment?

The standard rollout model pushes out tools, layers generic training on top, and assumes employees will work out how to change their own workflows. Most do not, and the reasons are well documented in behavioral science.

Status quo bias and loss aversion explain much of the resistance. Leaders tend to focus on what AI adds: innovation, speed, efficiency. Employees weigh what they might lose: expertise built over a career, job security, or confidence in their own output. Perceptyx research on AI readiness, drawn from more than 3,600 employees across North America and Europe, found 38% of employees are unclear on how AI will affect their roles and 53% fear bias or discrimination in AI-driven decisions. Resistance under those conditions is a normal human response to uncertainty, and more training does not resolve it, because the barrier was never a lack of information.

Resistance is the failure mode leaders can see. The one the workshop calls “horseless carriage thinking” — car prototypes that look like and are being used like carriages from the previous generation of technology — is harder to catch, because the employees producing it are daily users who register as adopted on every deployment dashboard. Reducing the fear behind resistance and redesigning the tasks behind unchanged output call for different interventions, so the diagnosis has to identify which population is which before any fix is selected.

What Is the COM-B Model and How Does It Diagnose Adoption Barriers?

COM-B is a behavioral science framework developed by Michie, van Stralen, and West in 2011. It holds that any behavior requires three components at once, and the workshop turns each into a practical diagnostic question:

• Capability: Can they do it? Do employees have the knowledge and skills to use AI effectively in their specific roles?
• Opportunity: Does the environment allow it? Do resources, processes, manager support, and time to experiment make the behavior possible?
• Motivation: Do they want to do it? Do employees have the confidence and drive to experiment proactively and iterate on what AI produces?

The diagnostic value comes from what most rollouts miss. Standard AI enablement programs address capability alone: deploy the tool, run the training, count the completions. When the real barrier sits in opportunity (a manager who has not created space to experiment) or motivation (an employee who fears the tool will erode their expertise), a better training program changes nothing. Diagnosing which component is blocking adoption, for which population, determines whether the right intervention is skill-building, workflow redesign, or manager coaching.

How Does COM-B Map to the People Insights Model?

COM-B on its own identifies the category of barrier. To act on it, the workshop maps each COM-B element onto a factor in the Perceptyx People Insights Model, the framework of 10 employee experience factors that connects survey items to specific behaviors and interventions:

• Capability maps to Growth & Development, measured through the behaviors of curiosity and sharing.
• Opportunity maps to Performance Enablement, measured through piloting and integrating AI into existing workflows.
• Motivation maps to Change & Innovation, measured through reflection and confidence.

The mapping described above accomplishes three things a standalone COM-B assessment cannot. It converts theory into benchmarkable survey data: the AI Adoption Pulse Survey that anchors the workshop uses 24 scaled items built to locate friction across the three domains. It gives HR and business leaders a shared vocabulary they already use for the rest of their listening program. And because every item ties to a behavior and an intervention pathway already established in the People Insights Model, the barriers identified in the survey point directly to a response instead of ending at a finding. A capability gap in sharing calls for a different intervention than a motivation gap in confidence, and the survey can tell those apart.

What Happens Inside the Activating AI Adoption Workshop?

The workshop is a two-hour facilitated session where stakeholders work directly with their own pulse survey data. Perceptyx consultants walk the group through the results using the COM-B and People Insights Model framework, identifying hot spots where friction is concentrated and bright spots where adoption is working and can be studied.

The most hands-on portion is the Nudge Lab, roughly 45 minutes where breakout groups design environmental nudges: small changes to the work environment that make the desired AI behavior the path of least resistance. Teams begin with empathy mapping, stepping into the perspective of a specific employee group to understand what that population sees, hears, and feels about AI tools. Working from a provided workbook and a menu of real world nudge examples mapped to the six AI adoption behaviors, groups then design interventions matched to their own data, such as changing default settings or building AI tools directly into an existing workflow, and define how they will measure whether each one worked. Because leaders and AI champions design the interventions themselves, they own the resulting plan.

The session closes with a roadmap across three horizons: immediate actions for the first few weeks, embedding changes into everyday capabilities and leadership behaviors, and the governance needed to scale and sustain progress.

How Does Behavior Change Continue After the Workshop Ends?

A two-hour session cannot change habits on its own, so the workshop connects to Activate, the behavior engine of the Perceptyx People Activation System. Six AI adoption behaviors have been added to the People Insights Model, and Intelligent Nudges tied to those behaviors deploy after the workshop through Microsoft Teams, Slack, email, and the mobile tools frontline employees already use. The research basis for this design is a 2024 field experiment by Haunstrup and Jensen, which found that training paired with just-in-time nudges produced durable behavior change measured eight months later, while training alone did not. Perceptyx data shows an average of 66% of managers engage with nudges, and teams receiving targeted nudges have seen up to a 12-point increase in related engagement scores within six months.

The full sequence runs baseline, workshop, reinforcement, and remeasurement. The pulse survey that opened the engagement can run again to show whether the friction it found has moved, which gives leadership an evidence chain from readiness data to intervention to measured change.

Ready to Diagnose Your Organization's AI Adoption Barriers?

Schedule a meeting with our team to learn how the Activating AI Adoption workshop can baseline your organization's readiness and turn the results into targeted action. For the research behind employee attitudes toward AI at work, read Beyond the Hype: Global Employee Perspectives on Generative AI.