The Adoption Gap

Why AI adoption stalls, and which factor actually predicts whether people keep using it.

Originally published: 5 October 2026
Last updated: 11 November 2026
DOI: 10.5281/zenodo.23306286

  • Brief Summary

    Most organizations assume AI adoption stalls because the rollout was poorly communicated, training was thin, or the tool was not accurate enough. We tested that assumption in a survey of 523 professionals whose organizations already use AI. Psychological trust predicts whether people use AI and intend to keep using it more strongly than any other category measured: about twice the weight of the next-strongest category and more than ten times the weight of change management. The pattern holds in healthcare, finance and insurance, and manufacturing and operations.

  • 523

    professionals surveyed across three sectors, all current users of AI tools.

  • 2×

    psychological trust's predictive weight versus the next-strongest category.

  • 10×+

    psychological trust's weight versus change management specifically.

  • 3 of 3

    sectors where psychological factors are the largest predictor of usage.

  • What Actually Predicts Usage

    Figure 1 compares the five categories we measured: change management, technical quality, incentives and culture, regulatory and compliance confidence, and psychological trust. Entered into one model that predicts usage and intent to keep using AI, psychological factors carry roughly double the weight of the next-strongest category (β = 0.391, against 0.182 for Technical Quality and 0.180 for Incentives & Culture). They carry more than ten times the weight of change management (β = 0.037), the lever most organizations pull first. Change management has no independent effect, and regulatory confidence is only marginal.

    Technical quality and incentives still matter. The point is the ranking: when all five explanations compete, trust wins.

    What Actually Predicts Usage
  • Trust Has Five Parts, and One of Them Does Not Matter

    Figure 2 breaks the psychological category into its five dimensions from the AI Trust Axis. Four predict usage on their own: feeling in control rather than overridden (Autonomy Safety, β = β = 0.165), the AI weighing the same factors the person would (Cognitive Alignment, β = 0.148), knowing what to do when an output cannot be trusted (Failure Recovery Intelligence, β = 0.136), and getting useful results with little effort (Interaction Effort, β = 0.101). The fifth, understanding how the AI reaches its outputs (Fairness Comprehension), shows no independent relationship to usage (β = −0.021, not significant). Trust in the outcome and the process appears to matter more than comprehension of the mechanism. Because the effect is spread across dimensions, no single transparency initiative or training module is likely to close the gap.

    Trust Has Five Parts, and One of Them Does Not Matter
  • The Pattern Holds in Every Sector

    As Figure 3 shows, psychological factors are the largest predictor in healthcare (β = 0.443), manufacturing and operations (0.427), and finance and insurance (0.310). What joins psychology differs. In healthcare, technical accuracy does, and incentives carry no independent weight. In manufacturing, incentives and culture are the second factor. Finance is the most multi-causal, with psychology, incentives, and technical quality all in play.

    The Pattern Holds in Every Sector
  • What This Means for Adoption Programs

    Invest in trust-building before communication campaigns. Treat trust as a precondition that lets accuracy and incentives pay off, not as one input among equals. Make override easy, visible, and normal, so people keep a felt sense of control. Build for error recovery as well as error prevention, so people know what to do when output looks wrong. And do not confuse explainability with trust: explaining how a model works may not be what is blocking adoption.

  • Download the Full Report

    The full report adds the five-category measurement table, a plain-language guide to reading the regression results, the behavioral theory behind each trust dimension, the full sector write-ups, and the methodology with responses to likely objections. It is free; enter your details to download it.

  • About the Study

    Respondents (n = 523; 84% in the United States, 9% in the United Kingdom, and 7% in Canada) rated agreement with thirteen statements across the five categories and two usage items, which were combined into one outcome. Sectors were healthcare (n = 166), finance and insurance (n = 158), and manufacturing and operations (n = 167); 32 respondents in other sectors are included in the overall model only. Coefficients show each category’s contribution with the other four held constant.