Skip to content

AI Adoption Framework

Why This Matters Now

Most organizations have already bought the tools. By 2025 roughly four in five had piloted generative AI and nearly 40% reported some deployment — yet MIT's review of more than 300 enterprise initiatives found that about 95% of organizations were getting zero return, with the deployed tools "primarily enhancing individual productivity, not P&L performance" (MIT NANDA, 2025). BCG reached the same place from a different sample: 74% of companies had yet to show tangible value from AI, and only about 4% had built capabilities generating significant value (BCG, 2024). A year later BCG's follow-up was blunter still — only 5% of companies were achieving value at scale while 60% were achieving none at all (BCG, 2025).

~95% of organizations are getting zero return from enterprise generative AI — the tools enhance individual productivity but do not move the P&L (MIT NANDA, 2025).

This is not a model-capability gap. The capability has been bought and, in most places, switched on. The gap is between access and adoption — between a license that exists and a workflow that has actually changed because a person changed how they work. This page is the organizing argument for that human layer. It defines what adoption requires beyond rollout, the mindset shift from AI-as-tool to AI-as-collaborator, the trust dynamics that determine whether people lean on AI too little or too much, the specific resistance patterns AI provokes, and why culture is the binding constraint that gates every other track in this framework. The companion pages tell you how to run an adoption program and how to score where you stand.

Adoption Is Not Access

The single most expensive error in AI transformation is treating procurement as progress. Seats are provisioned, a launch email goes out, a usage dashboard ticks up in the first week — and leadership marks the initiative delivered. But the dashboard measures access, and value comes only from sustained use that reshapes work.

The distance between the two is large and well-documented. McKinsey reports that 78% of organizations now use AI in at least one business function (McKinsey, 2024) — but only 39% can attribute any enterprise-level EBIT impact to it, and most of those put the contribution below 5% of EBIT (McKinsey, 2025). The same research isolates the mechanism: only about 21% of organizations using generative AI had redesigned any workflow around it, and of the dozens of attributes tested, workflow redesign showed the strongest correlation with bottom-line impact (McKinsey, 2025). Tool rollout without workflow change is the default failure mode, and it is overwhelmingly the common case.

So "adoption" in this framework means something specific:

  • Access — the tool is provisioned and reachable. Necessary, trivial, and almost never the constraint.
  • Activation — a person has used it for real work at least once, past the demo.
  • Habitual use — the tool is part of how recurring work gets done, not a novelty pulled out occasionally.
  • Workflow integration — the surrounding process has been redesigned so the tool's output flows into the next step rather than being bolted onto an unchanged sequence.

Value lives at the last two levels. Most measurement stops at the first. The job of an adoption effort is to move people up this ladder — and that is a human and cultural problem long before it is a technical one. BCG frames the resource split as roughly 10% of the effort on algorithms, 20% on technology and data, and 70% on people and process (BCG, 2024). The 70% is what this track is about.

From Tool to Collaborator: The Mindset Shift

Earlier enterprise software automated deterministic tasks: the same input produced the same output, and the user's mental model was "operate the machine." Generative AI breaks that model. It is probabilistic, conversational, occasionally wrong, and improves with how it is directed — closer to delegating to a capable but fallible junior colleague than to running a report. Adoption stalls when people carry the old mental model into the new tool: they expect deterministic correctness, get a plausible-but-wrong answer, and conclude the tool is unreliable.

The shift that has to happen is from AI as tool to AI as collaborator — from "operate it and trust the output" to "direct it, evaluate its output, and iterate." That reframing is the precondition for everything else, and it is fundamentally about calibrated trust: trusting AI for what it is good at, verifying where it is weak, and knowing the difference.

The two failure modes of trust

Trust in AI fails in both directions, and an adoption strategy has to manage both.

Under-trust (algorithm aversion). People discount or abandon a capable system, often after seeing it make a single error — even when it outperforms the human alternative. The foundational experiments showed that people lose confidence in an algorithm faster than in a human after seeing each make the same mistake, and that giving users even a small ability to adjust an imperfect algorithm sharply increases their willingness to use it (Dietvorst et al., 2015). Under-trust shows up as the employee who tried a chatbot once, caught a wrong answer, and never returned.

Over-trust (automation bias). The opposite and equally damaging error: treating the automated output as authoritative and dropping the vigilance a human colleague would receive. Decades of human-factors research document automation bias and the complacency that accompanies it — users commit errors of omission (missing what the system missed) and commission (following a wrong recommendation against contrary evidence) (Parasuraman & Manzey, 2010). The current data shows this is not hypothetical: 66% of employees report having relied on AI output at work without critically evaluating it, and 56% say they have made mistakes in their work because of AI (KPMG, 2025; KPMG, 2025).

66% have relied on AI output without critically evaluating it; 56% have made work mistakes because of AI — over-trust is as much an adoption risk as under-trust (KPMG, 2025; KPMG, 2025).

The baseline disposition is wary: only 46% of people globally say they are willing to trust AI systems (KPMG, 2025). The goal of an adoption program is not to maximize trust — it is to calibrate it: raise it where the tool is reliable and the human is the bottleneck, and hold it down where the output needs verification. A culture that punishes every AI error pushes people into under-trust and abandonment; a culture that celebrates AI uncritically pushes them into over-trust and unmanaged risk. Both destroy value.

The Resistance Patterns AI Provokes

Resistance to AI is not generic change resistance. It has specific, repeatable shapes, and naming them is the first step to addressing them. The prevailing sentiment is anxious: 52% of US workers say they feel worried about future AI use in the workplace, against just 36% who feel hopeful (Pew Research Center, 2025).

Fear of job loss. The most cited barrier, and partly rational. 18% of US employees think it likely their job will be eliminated within five years because of AI or automation — rising to 23% at organizations that have already adopted AI (Gallup, 2026). An employee who believes a tool exists to replace them has no incentive to make it work well, and every incentive to quietly withhold the knowledge that would.

Fear of looking incompetent — or like a cheat. A subtler and arguably larger brake. 48% of desk workers say they would feel uncomfortable admitting to their manager that they had used AI for a common task; the reasons they give are that it feels like cheating (47%), that they fear being seen as less competent (46%), and that they fear being seen as lazy (46%) (Slack, 2024). Microsoft's data echoes it: 52% of workplace AI users are reluctant to admit using AI for their most important tasks, and 53% worry it makes them look replaceable (Microsoft, 2024).

Distrust of outputs. The algorithm-aversion pattern above, expressed as a reason not to bother — "I'd have to check it anyway, so it's faster to do it myself." Often legitimate for a specific weak use case, but frequently over-generalized from one bad experience to the whole tool.

Loss of autonomy and control. Knowledge workers resist tools they experience as surveillance or as dictating how to work rather than helping them work. Adoption framed as mandate-and-monitor reliably backfires.

Resistance is not uniform — map it

These patterns are not evenly distributed across a workforce. Rogers' diffusion-of-innovations model remains the cleanest way to see this: any population splits into innovators (2.5%), early adopters (13.5%), an early majority (34%), a late majority (34%), and laggards (16%) (Rogers, 2003). The strategic error is to design the whole program for the enthusiasts — the ~16% who would adopt anything — and then be surprised when momentum stalls at the majority. The hardest and most valuable transition is from the early adopters to the pragmatic early majority, who adopt on proof rather than novelty. An adoption strategy that cannot cross that gap reaches the willing and no further.

Shadow AI: resistance and adoption at the same time

The most revealing pattern is that suppressed adoption does not disappear — it goes underground. 78% of AI users bring their own AI tools to work rather than wait for sanctioned ones (Microsoft, 2024), and roughly half of employees use "shadow AI" tools their company did not provide (Software AG, 2024). Crucially, prohibition makes this worse, not better: inappropriate AI use — such as putting sensitive company data into public tools against policy — was found to be more common among employees in organizations that ban generative AI (67%) than in those with no such ban (33%) (KPMG, 2025). Shadow AI is simultaneously evidence of strong demand and a governance liability — the same employees hiding their use are often pasting sensitive data into consumer tools. It cannot be banned away; it has to be channeled into sanctioned tools and clear acceptable-use guidance (see AI Policy & Acceptable Use).

Leadership Modeling and Psychological Safety

If the dominant resistance pattern is fear of looking incompetent or like a cheat, the dominant lever is making it safe to use AI openly — and that is set at the top.

The clearest empirical signal is from Slack: workers who feel comfortable disclosing their AI use are 67% more likely to have actually used AI for work than those who feel they have to hide it (Slack, 2024). Disclosure comfort is not a soft nicety; it tracks directly with usage. And much of the workforce has been left in the dark: 45% of desk workers say they lack explicit permission or guidelines to use AI, and 30% have had no AI training at all (Slack, 2024). Silence from leadership is not neutral — it is read as disapproval, which pushes use into the shadows.

This is the organizational analogue of psychological safety — "a shared belief held by members of a team that the team is safe for interpersonal risk taking" — which research has long shown to be the strongest enabler of team learning behavior (Edmondson, 1999), and which Google's large internal study identified as by far the most important dynamic of effective teams (Google re:Work, 2015). Using a new, fallible tool in front of colleagues is interpersonal risk-taking. Without safety, people won't experiment, won't admit when AI helped, and won't surface what went wrong — which is exactly the information an adoption program needs.

Leadership modeling is what creates that safety. Leaders who visibly use AI, talk about where it failed them, and treat experimentation as expected rather than suspicious give everyone else permission. The gap here is real: 79% of leaders agree their company must adopt AI to stay competitive, but 60% worry their organization lacks a plan or vision to implement it (Microsoft, 2024). Conviction without a visible, modeled plan produces anxious employees and stalled adoption.

Why Culture Gates Every Other Track

This framework has tracks for strategy, governance, data readiness, technology and platform, and workflow optimization. Each is necessary. None pays off without adoption, because every one of them ultimately routes through a human being changing what they do.

  • A sound strategy that names the right priorities still requires people to act on them. Strategy decks do not change behavior; adoption does.
  • Governance policies only reduce risk if people follow them — and the shadow-AI data shows that policy without adoption produces worse compliance than no policy at all (KPMG, 2025).
  • Data and platform investments build capability that sits idle until someone uses it. The 95% zero-return finding is, in large part, the cost of capability that was deployed but never adopted (MIT NANDA, 2025).
  • Workflow optimization is where value is actually realized — and it depends entirely on people willing to let a workflow be redesigned around AI. The fact that only ~21% of organizations have redesigned any workflow is not a tooling failure; it is an adoption failure (McKinsey, 2025).

This is why culture is the gate, not a parallel afterthought. An organization can be ahead on every technical track and still capture no value if its people don't trust, don't use, or actively route around the tools. Conversely, the organizations in the 5% capturing value at scale (BCG, 2025) are not distinguished primarily by better models — those are increasingly commodity — but by having built the trust, safety, and changed behavior that let capability translate into outcomes.

The practical implication is sequencing: adoption is not the celebration at the end of a rollout. It is a workstream that runs alongside every other track from the start, with its own owner, plan, and metrics. The next two pages make that concrete — designing the program and measuring adoption maturity.

Key Takeaways

  • Access is not adoption. ~95% of organizations get zero return because tools are deployed but workflows never change; value lives in habitual use and workflow integration, not provisioned seats (MIT NANDA, 2025; McKinsey, 2025).
  • Manage trust in both directions. Under-trust (algorithm aversion) causes abandonment; over-trust (automation bias) causes unmanaged error. The goal is calibration, not maximization (Dietvorst et al., 2015; Parasuraman & Manzey, 2010; KPMG, 2025).
  • Name the resistance. Fear of job loss, of looking incompetent, of looking like a cheat, distrust of outputs, and loss of autonomy are distinct patterns with distinct responses — and they are not uniform across the adoption curve (Slack, 2024; Pew, 2025; Gallup, 2026; Rogers, 2003).
  • Safety and modeling are the primary levers. Disclosure comfort tracks a 67% higher likelihood of actual use; leaders who model and permit AI use create the psychological safety that converts demand into open, governable adoption (Slack, 2024; Edmondson, 1999).
  • Culture gates everything. Strategy, governance, data, platform, and workflow each route through human behavior — adoption is the binding constraint on the value all of them are meant to produce.

Sources

  • MIT Project NANDA — The GenAI Divide: State of AI in Business 2025, 2025. Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact. View source · verified 2026-06-20 · ⚠ secondary mirror
  • BCG — Where's the Value in AI?, 2024. Only 26% of companies have developed the necessary set of capabilities to move beyond proofs of concept and generate tangible value (4% cutting-edge plus 22% advanced). View source · verified 2026-06-20 · primary
  • BCG — The Widening AI Value Gap, 2025. Only about 5% of companies (the 'future-built') are generating value at scale, while 60% of companies are laggards with little or no value; 35% are in between. View source · verified 2026-06-20 · primary
  • McKinsey — The State of AI: How Organizations Are Rewiring to Capture Value, 2024. 78% of organizations now use AI in at least one business function (up from 72% in early 2024 and 55% in 2023). View source · verified 2026-06-22 · primary
  • McKinsey — The State of AI, 2025. About 39% of organizations report any enterprise-level EBIT impact from AI; most of those say less than 5% of EBIT is attributable to AI use. View source · verified 2026-06-20 · primary
  • McKinsey — The State of AI, 2025. 21% of respondents reporting gen AI use say their organizations have fundamentally redesigned at least one workflow; redesigning workflows has the biggest effect on an organization's ability to see EBIT impact from gen AI. View source · verified 2026-06-20 · primary
  • BCG — Where's the Value in AI?, 2024. AI leaders follow the rule of putting 10% of their resources into algorithms, 20% into technology and data, and 70% into people and processes. View source · verified 2026-06-20 · primary
  • Dietvorst, Simmons & Massey — Algorithm Aversion: People Erroneously Avoid Algorithms After Seeing Them Err (Journal of Experimental Psychology: General, 144(1), 114-126), 2015. People more quickly lose confidence in algorithmic than human forecasters after seeing them make the same mistake; giving people even a slight ability to modify an imperfect algorithm substantially increases their willingness to use it. View source · verified 2026-06-20 · primary
  • Parasuraman & Manzey — Complacency and Bias in Human Use of Automation: An Attentional Integration (Human Factors, 52(3), 381-410), 2010. Automation-induced complacency and automation bias produce errors of omission (missing events the automation does not flag) and commission (following an automated recommendation despite contrary evidence). View source · verified 2026-06-20 · primary
  • KPMG & University of Melbourne — Trust, Attitudes and Use of AI: A Global Study 2025, 2025. Two-thirds of employees (66%) report having relied on AI output at work without critically evaluating the information it provides. View source · verified 2026-06-20 · primary
  • KPMG & University of Melbourne — Trust, Attitudes and Use of AI: A Global Study 2025, 2025. Over half (56%) report they have made mistakes in their work because of AI use. View source · verified 2026-06-20 · primary
  • KPMG & University of Melbourne — Trust, Attitudes and Use of AI: A Global Study 2025, 2025. Only 46% of people globally are willing to trust AI systems. View source · verified 2026-06-20 · primary
  • Pew Research Center — U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace, 2025. About half of workers (52%) say they are worried about the future impact of AI use in the workplace; 36% say they feel hopeful; about a third (32%) say it will lead to fewer opportunities for them. View source · verified 2026-06-20 · primary
  • Gallup — Rising AI Adoption Spurs Workforce Changes, 2026. 18% of all U.S. employees say it is very or somewhat likely their job will be eliminated within the next five years due to AI or automation; among employees in organizations that have adopted AI, that share rises to 23%. View source · verified 2026-06-20 · primary
  • Slack Workforce Lab — Fall 2024 Workforce Index, 2024. Nearly half (48%) of all desk workers would be uncomfortable admitting to their manager that they used AI for common workplace tasks; reasons include feeling like using AI is cheating (47%), fear of being seen as less competent (46%), and fear of being seen as lazy (46%). View source · verified 2026-06-20 · primary
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. 52% of people who use AI at work are reluctant to admit using it for their most important tasks; 53% worry that using AI on important work tasks makes them look replaceable. View source · verified 2026-06-20 · primary
  • Everett M. Rogers — Diffusion of Innovations (5th ed.; ISBN 978-0743222099), 2003. Adopter categories: innovators 2.5%, early adopters 13.5%, early majority 34%, late majority 34%, laggards 16%, obtained by partitioning the normal adoption distribution at the mean plus or minus one and two standard deviations. View source · verified 2026-06-20 · ⚠ secondary mirror
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. 78% of AI users are bringing their own AI tools to work (BYOAI). View source · verified 2026-06-20 · primary
  • Software AG — Shadow AI study, 2024. Half of all employees are shadow-AI users, relying on unsanctioned or ad-hoc AI tools their company did not provide. View source · verified 2026-06-20 · ⚠ secondary mirror
  • KPMG & University of Melbourne — Trust, Attitudes and Use of AI: A Global Study 2025, 2025. Using AI in ways that contravene organizational policies is most common among employees who report their organization has banned generative AI (67%), compared with those in organizations without such policies (33%). View source · verified 2026-06-20 · primary
  • Slack Workforce Lab — Fall 2024 Workforce Index, 2024. Those who are comfortable sharing that they used AI for work tasks are 67% more likely to have used AI for work than those who would not be comfortable admitting AI use. View source · verified 2026-06-20 · primary
  • Slack Workforce Lab — Fall 2024 Workforce Index, 2024. Close to half of workers (45%) do not have explicit permission to use AI; 30% say they have had no AI training at all, including no self-directed learning or experimentation. View source · verified 2026-06-20 · primary
  • Amy C. Edmondson — Psychological Safety and Learning Behavior in Work Teams (Administrative Science Quarterly, 44(2), 350-383), 1999. A shared belief held by members of a team that the team is safe for interpersonal risk taking; team psychological safety is associated with learning behavior, which mediates between psychological safety and team performance. View source · verified 2026-06-20 · primary
  • Google re:Work — Project Aristotle: Understand Team Effectiveness, 2015. Psychological safety was far and away the most important of the five dynamics we found; it is the underpinning of the other four. View source · verified 2026-06-20 · primary
  • Microsoft & LinkedIn — 2024 Work Trend Index Annual Report, 2024. 79% of leaders agree their company needs to adopt AI to stay competitive, but 60% worry their organization's leadership lacks a plan and vision to implement it. View source · verified 2026-06-20 · primary