Making AI Adoption Stick: The Human-AI System

Part 1 of a 5-part series: The human AI system.

The neuroscience of the implementation gap - and why fixing it starts with humans, not technology

Your organisation has an AI strategy. Leadership is aligned, the business case is approved, and the roadmap has been communicated. The vendor is onboarded. The training is being rolled out, and the pilot results were promising. So why aren’t people actually using it?

This question is keeping CHROs, COOs and Chief Digital Officers awake right now. Not because they lack ambition or resources, but because they are discovering - often too late - that their AI strategy was built on a flawed assumption: that if you build it, they will adopt. Many people won’t. And the reason is not what most strategy documents are willing to say.

The reason is the brain.

If the technology is working - and it largely is - what exactly is failing? The answer, in organisation after organisation, is the same: the human layer nobody modelled in the strategy.

This is the first piece in a new series I am calling The human AI system, an extension of my earlier AI anxiety series that moves from individual psychology into organisational strategy.

This series asks why organisations keep failing to act on what they know. The answer is that they are trying to solve a neuroscience-based problem with a project-management toolkit.

The implementation gap is real, and it is getting worse

The numbers are sobering. MIT’s Project NANDA, in The GenAI Divide: State of AI in Business 2025, found that roughly 95% of organisations were seeing no measurable return on their generative-AI investments within the study’s measurement window (MIT Project NANDA, 2025). BCG’s research tells a complementary story: only about 5% of companies are capturing value from AI at scale, while close to 60% report little or no material return despite significant investment - and BCG attributes that gap primarily to human and organisational factors rather than to the technology itself (BCG, 2025, 2026a). And S&P Global Market Intelligence found that 42% of companies abandoned most of their AI initiatives in 2025, up from just 17% the year before (S&P Global Market Intelligence, 2025).

These are not technology failures. The algorithms are working and the models are more capable than ever. The infrastructure is in place. What is failing is the human layer underneath the technology.

Leaders consistently tell me that AI adoption challenges have the capacity to do significant damage to organisational performance - not because of technical problems, but because of people problems that no one modelled in the original strategy. The keys lie in bridging the gaps in awareness about organisational strategy and, related to this, trust, and finally the actual behaviours that demonstrate healthy adoption. This should also include a feedback loop that helps spotlight, celebrate and reinforce those behaviours, creating new habits that become part of the organisational DNA.

Each of these gaps has a name in the strategy deck: communication challenge, change resistance, governance issue. But these labels obscure the mechanism. What is actually happening in each case is a predictable neurological response to perceived threat - and until organisations understand that, they will keep addressing the symptom while the root cause compounds.

The brain was not designed for this

To understand why AI adoption fails at the human layer, we need to understand what is happening inside the brain when people are asked to change how they work - particularly when the change is perceived as a threat to identity, expertise or status.

The threat response is not optional

The human brain’s threat-detection system, centred in the amygdala, does not reliably distinguish between physical danger and social or professional threat. Neuroscience has shown that social rejection and exclusion recruit much of the same neural circuitry as physical pain (Eisenberger et al., 2003). Job displacement, obsolescence and loss of expert status activate that response just as more visceral dangers do. The result is a cascade of stress hormone and neurotransmitter surges that, as Arnsten (2009) documents, impairs prefrontal function - narrowing thinking, reducing cognitive flexibility and, critically, making people less willing to experiment, collaborate or admit uncertainty.

This is not weakness or resistance. It is biology. And it is the single most underestimated variable in every AI implementation plan I have reviewed.

What this means in practice is that when you announce an AI implementation, a significant proportion of your workforce immediately enters a low-grade threat state. They may not say so. They may smile in the town hall. But their neural architecture has already begun to protect them - through avoidance, information hoarding, surface compliance and what researchers describe as maladaptive coping behaviours: using AI tools covertly, withholding information about errors, or finding workarounds that look like adoption but produce none of the intended value.

The three needs AI adoption must meet

Decades of motivational psychology tell us that human beings have three fundamental psychological needs that, when threatened, produce disengagement and resistance: Competence (the need to feel capable), Autonomy (the need to feel in control of one’s actions) and Relatedness (the need to feel connected and valued within a social group) (Deci & Ryan, 2000).

Most AI implementations threaten all three simultaneously. Competence is threatened because the tools are new and employees fear looking incompetent during the learning curve. Autonomy is threatened because AI is often introduced top-down, with no co-design and limited individual choice. Relatedness is threatened because the adoption timeline is rarely consistent, meaning some colleagues are visibly ahead while others are still struggling - fracturing team cohesion.

The organisation sends the message: adopt this or be left behind. The brain hears: you are already behind, you have no control, and your peers and bosses are noticing.

This is not a communication problem. It is a motivational-architecture problem that requires a different kind of strategy to solve.

The double-bind of middle management

Of all the human-layer challenges in AI implementation, none is more consequential than the position of middle management.

Middle managers are typically the people asked to translate AI strategy into team-level behaviour. They are expected to model enthusiasm, answer questions, coach adoption and maintain performance - all while privately navigating their own deep uncertainty about what AI means for their role and career.

This is a kind of neurological self-reinforcing trap. How can you effectively help regulate your team’s anxiety about a threat you have not processed yourself? Given that anxiety is socially contagious (Barsade, 2002), a leader who is struggling with their own threat response will transmit that state to their team through micro-signals - tone, pace, the questions they avoid, the certainty they perform. In my earlier work on AI anxiety as social contagion, I documented how a single team member can shift the emotional tone of an entire group. At the middle-management level, that effect is amplified by authority.

The middle manager threat profile

Gartner (2024) predicted that, through 2026, 20% of organisations will use AI to flatten their structures, eliminating more than half of current middle-management positions. The people being asked to drive AI adoption are the ones most existentially threatened by it. Research identifies two characteristic defence responses: Analysis Paralysis (endless proof-of-concept requests as a shield against failure) and Knowledge Hoarding (protecting expertise as the perceived last line of job security). Neither behaviour looks like resistance. Both are indistinguishable from diligence or prudence. This is what makes the middle layer so difficult to diagnose without the right measurement tools.

The strategic implication is significant. If your AI implementation plan asks middle managers to lead change they are not psychologically equipped to lead, your strategy is not failing because of the technology. It is failing because of an unexamined assumption about human capacity under threat.

The C-suite is focusing on a different problem

There is a final dimension to this gap that most organisations are not willing to name directly: the extent to which your people have a sense of autonomy or authorship.

Senior leaders typically have significant input into AI strategy decisions, and BCG now finds CEOs increasingly own those decisions outright (BCG, 2026b). They participate in vendor selection, shape the rollout timeline and understand the business rationale. This gives them a sense of authorship, which is one of the most powerful buffers against threat. When you helped create the change, your brain encodes it as something you are doing to the world rather than something the world is doing to you.

Frontline employees and most middle managers had no part in the AI decision. They are recipients of a strategy they did not shape, about tools they did not choose, on a timeline they had little hand in setting. Their neural threat response is structurally predictable. Yet the standard implementation plan treats them as if their psychology should mirror the executive team’s enthusiasm.

Leaders are optimistic about AI. The people actually using the tools are sceptical.
The most revealing pattern here is the gap between how leaders imagine their people feel and how they actually feel. In research reported by Workday, 76% of executives believed their employees were enthusiastic about AI, while just 31% of those employees actually were - and the more senior the leader, the more they overestimated that enthusiasm (Workday, 2026). That gap is not bridged by a better slide deck or a more enthusiastic launch event. It is bridged by rebuilding the conditions the brain requires to feel safe enough to learn.

What the brain actually needs to adopt

The good news is that the neuroscience provides the foundations for practical solutions. The science offers considerable and precise insight into the conditions the human brain requires in order to learn new behaviours under uncertainty.

There are some core conditions that consistently produce genuine adoption rather than surface compliance. They are not complex, but they are consistently absent from standard implementation plans.

Acknowledge before you activate
The single most underinvested step in AI implementation is the one that happens before deployment: explicitly naming and normalising the anxiety. Absence of conversation does not mean absence of anxiety. It means the anxiety is finding another outlet - typically in avoidance, sabotage or shadow use.

Leaders who go first - who openly acknowledge that this is uncertain, that they do not have all the answers, and that some of what people are feeling is appropriate and expected - do not weaken adoption. They accelerate it, because they remove the social cost of visible uncertainty. When it feels safe to struggle, people are more likely to try.

Rebuild the three psychological needs
Effective AI adoption strategies are explicitly designed around competence, autonomy and relatedness. This means:

Competence: structured, low-stakes experimentation before performance expectations are attached to AI use. Sandboxed learning environments. Explicit permission to be a beginner.
Autonomy: co-design at the team level. Giving people genuine choices about how they integrate AI into their workflows, not just whether. Participation in identifying use cases.
Relatedness: peer cohorts, collective learning rhythms, and visible adoption as a shared experience rather than an individual performance assessment.

Address the middle layer directly and separately
Middle managers need their own activation pathway - not a version of the frontline programme with a leadership badge on it. They need specific work on their own threat response, their professional identity in an AI-augmented environment, and the self-regulation skills to lead their teams without transmitting their own unprocessed anxiety.

With reported levels of burnout are increasing year upon year, building manager psychological resources through systems, role clarity and other organisational structures that optimise trust (rather than placing it all on the individual to simply "toughen up") is also crucial to the Human Operating System. Gallup’s State of the Global Workplace 2026 reports that the deepening global slide in engagement and wellbeing is being driven not by frontline workers but by their managers, whose own burnout is accelerating: manager engagement fell five points in a single year, from 27% to 22% - the steepest single-year drop Gallup has recorded - with younger and female managers hit hardest (Gallup, 2026). AI is not arriving into a resilient, stable workforce. It is arriving into one that is already running close to capacity. Middle managers are the load-bearing walls of that structure.

Measure what actually matters
Most organisations measure AI deployment: licences activated, tools installed, training modules completed. These metrics are the organisational equivalent of measuring learning by attendance. They tell you nothing about whether the human layer has actually shifted.

What needs to be measured is the psychological infrastructure beneath the adoption behaviour: psychological safety around AI use, identity-threat levels by team and function, the gap between stated intention and actual behaviour, and the quality of leader modelling at every tier. Without this data, you are navigating a human problem with a technology dashboard.

The question every senior leadership team should be asking

I have spent years working with some of the world’s most capable organisations on the human dimensions of leadership, trust and change. What I observe in AI implementation projects right now is a familiar pattern: organisations that are technically sophisticated and strategically ambitious, but that are systematically underinvesting in the variable most likely to determine whether their investment pays off.

The question every senior leadership team should be asking is not just “How do we deploy AI?” It must move quickly on to “Are our people behaviourally equipped to adopt it?”

Those are different questions. They require different answers. And they require different expertise.

The organisations that will lead in the AI era are not the ones with the most sophisticated tools. They are the ones that invest as seriously in their Human Operating System as they do in their technology stack. That means diagnosing the human layer before deployment, activating the right conditions for genuine adoption, and measuring the psychological outcomes - not just the technical ones - over time.

Up next in this series:

Overcoming the middle manager paradox: Helping the people asked to lead AI adoption to adopt, role model and actually lead
Shadow AI and what it is really telling you: The behavioural signal organisations are misreading
Building the Human Operating System: A practical framework for AI-ready organisations
Measuring what matters: The metrics that predict genuine adoption (and the ones that don’t)

References

Arnsten, A. F. T. (2009). Stress signalling pathways that impair prefrontal cortex structure and function. Nature Reviews Neuroscience, 10(6), 410-422. https://doi.org/10.1038/nrn2648

Barsade, S. G. (2002). The ripple effect: Emotional contagion and its influence on group behaviour. Administrative Science Quarterly, 47(4), 644-675. https://doi.org/10.2307/3094912

BCG. (2025). The widening AI value gap: Build for the future 2025 global study. Boston Consulting Group. https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap

BCG. (2026a). AI transformation is a workforce transformation. Boston Consulting Group. https://www.bcg.com/publications/2026/ai-transformation-is-a-workforce-transformation

BCG. (2026b). AI radar 2026: As AI investments surge, CEOs take the lead. Boston Consulting Group. https://www.bcg.com/publications/2026/as-ai-investments-surge-ceos-take-the-lead

Deci, E. L., & Ryan, R. M. (2000). The “what” and “why” of goal pursuits: Human needs and the self-determination of behaviour. Psychological Inquiry, 11(4), 227-268. https://doi.org/10.1207/S15327965PLI1104_01

Eisenberger, N. I., Lieberman, M. D., & Williams, K. D. (2003). Does rejection hurt? An fMRI study of social exclusion. Science, 302(5643), 290-292. https://doi.org/10.1126/science.1089134

Gallup. (2026). State of the global workplace: 2026 report. Gallup. https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx

Gartner. (2024). Gartner top strategic predictions for 2025 and beyond: Riding the AI whirlwind [Press release]. Gartner, Inc. https://www.gartner.com/en/newsroom/press-releases/2024-10-22-gartner-unveils-top-predictions-for-it-organizations-and-users-in-2025-and-beyond

MIT Project NANDA. (2025). The GenAI divide: State of AI in business 2025. Massachusetts Institute of Technology. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf

S&P Global Market Intelligence. (2025). Voice of the enterprise: AI & machine learning, use cases 2025. S&P Global. https://www.spglobal.com/market-intelligence/en/news-insights/research/2025/10/generative-ai-shows-rapid-growth-but-yields-mixed-results

Workday. (2026). How the AI trust gap slows adoption (and what to do about it). Workday. https://www.workday.com/en-us/perspectives/ai/how-ai-trust-gap-slows-adoption.html*

Making AI adoption stick: The human AI system