Your privacy choices

    Necessary storage supports report access and site operation. Optional Google measurement and advertising stays off unless you accept. Forms, reports, video and booking work without it. Privacy Policy

    May 14, 202611 min readUpdated October 14, 2026Dariusz Ambroziak

    The AI Adoption Paradox: Why Investment Is Soaring While Transformation Stalls

    In McKinsey's January 2025 survey, 92% planned to increase AI investment over three years and 1% self-reported full maturity. Investment plans and maturity are different measures from realized value.

    The 10-20-70 rule of AI transformation: 10% algorithms, 20% technology and data, 70% people and process

    In McKinsey's January 2025 'Superagency in the Workplace' survey, 92% of companies planned to increase AI investment over the following three years, while 1% of leaders described their organizations as fully mature in AI deployment. These are investment intentions and self-reported maturity, not measures of realized value. Low full maturity does not mean that the other organizations receive no value.

    Welcome to the collaboration.tech blog. In this first post we show why AI transformation is, first and foremost, a psychological, organizational, and human challenge — and only after that a technological one. The January 2025 McKinsey figures describe investment plans and self-reported maturity, not observed revenue growth.

    Three numbers that change how we think about AI

    McKinsey's 'Superagency in the Workplace' report (January 2025) distinguishes planned investment from self-reported full maturity. Low full maturity does not mean no value. In numbers:

    • 92% of companies planned in the January 2025 survey to increase AI investment over the following three years
    • 1% of leaders in the January 2025 survey described their organizations as fully mature in AI deployment
    • 47% of C-suite leaders believe their company is moving too slowly on AI

    Boston Consulting Group, in its 'The Widening AI Value Gap' report (dated 30 September 2025), added an equally striking financial picture at the time:

    • BCG, 30 September 2025: 5% were classified 'future-built'. In the 30 September 2026 Applied AI Index, 7.5% are future-built and another 41% are scaling. Neither classification implies that all other companies receive no value.
    • BCG’s historical 30 September 2025 snapshot classified 60% as laggards reporting minimal or no financial returns; this is not a current universal failure rate.
    • BCG's September 2025 survey reported that AI leaders expected twice the revenue increase and 40% greater cost reductions than laggards in AI application areas, not realized whole-market outperformance

    This is not a technology gap. It's an organizational one.

    The 10-20-70 rule: where AI value actually comes from

    BCG has articulated one of the most important principles of AI transformation today. It sounds simple — but its implications run deep:

    10% of effort on algorithms. 20% on technology and data. 70% on people and processes.
    Bar chart visualising the 10-20-70 rule: 10% algorithms, 20% technology and data, 70% people and process
    The 10-20-70 rule of AI transformation — adapted from BCG, 'The Widening AI Value Gap' (2025).

    Read that twice. Seventy percent. The bulk of the work in an AI transformation is work on humans — on motivation, trust, capability building, team design, decision-making structures, and organizational culture. Not on models, and not on cloud infrastructure.

    BCG's 10-20-70 guideline is an allocation of effort, not evidence that most firms spend in an inverted ratio. McKinsey's historical 2025 self-assessment of 1% maturity does not imply that 99% receive no value. In the August 2026 State of AI survey, 37% of respondents report some EBIT impact and about 6% qualify as high performers: different measures, not a maturity trend from 1% to 6%.

    The 'jagged frontier': when AI helps, and when it hurts

    In 2023, Harvard Business School and BCG conducted one of the most rigorous studies to date on AI's impact on knowledge work. 758 BCG consultants (about 7% of the firm's global workforce) performed realistic tasks either with or without access to GPT-4.

    For tasks inside the so-called 'frontier of AI capability' (Dell'Acqua et al., 2026), the results were striking:

    • +12.2% more tasks completed
    • +25.1% faster completion
    • +40% higher quality

    But there's a second finding that rarely gets discussed out loud. On the specific outside-the-frontier task in the study — a task GPT-4 isn't yet competent at — consultants using GPT-4 produced 19 percentage points fewer correct answers than the group working without AI at all (Dell'Acqua et al., 2026). On that task, artificial intelligence didn't just fail to help. It actively hurt.

    This 'jagged technological frontier' is one of the most important concepts for anyone deploying AI. Artificial intelligence isn't uniformly good or uniformly bad. It has patchy competence — and it demands judgment from humans about when to trust it and when to keep distance. That's a capability you can't buy with a license. You have to build it inside the organization.

    Trust, psychological safety, and alienation

    This is where it gets interesting — and where organizational psychology enters the picture in earnest.

    Trust in AI erodes over time. Initial excitement about a tool gives way to disappointment as people discover its limitations (Schmutz et al., 2024). That's a natural psychological process — but for organizations it means the 'wow effect' of the first few weeks fades, and then the real work begins: calibrating expectations, designing processes, building respect both for what AI can do and for where it falls short.

    What's more, research shows that adding AI to a team often reduces coordination, communication, and trust between people (Schmutz et al., 2024). Hybrid teams (Human-AI Teams, HATs) counterintuitively perform worse than teams without AI. Why? Because people don't know how to collaborate with an agent that has no intuition, doesn't remember yesterday's agreement, and doesn't feel the consequences of an error. This isn't a lack of training on the tool. It's a missing new mental model of collaboration.

    Bezrukova et al. (2023) propose, as part of their conceptual model of AI and groups, an even more counterintuitive hypothesis: that when teams with a positive attitude toward AI are mandated to use it, their level of collaboration may decline — while teams with negative attitudes receiving the same mandate may paradoxically see collaboration rise. These are theoretical propositions, not empirical findings, and they can't be fixed with better technology alone. They point to the need for understanding psychology — autonomy, attitude congruence within a group, and power dynamics.

    One small but important note: in measurements like these, the average attitude of a team tells you almost nothing. What matters is the distribution and the congruence. A team in which everyone sits at 6/10 behaves completely differently from a team in which half sit at 10 and half at 2 — even though the average is identical.

    And this brings us back to psychological safety — the concept made famous by Amy Edmondson at Harvard Business School. Without it, employees won't experiment with AI. They won't speak up when AI is wrong. They'll hide their doubts. And the entire technology investment will turn out to be — a costly empty gesture. A 3-wave observational study of 381 South Korean workers found that AI adoption was associated with lower psychological safety, which in turn was associated with higher self-reported depressive symptoms (Kim, Kim & Lee, 2025) — an association based on self-report, not a clinical diagnosis, and not evidence of causation. This is not an add-on to a 'soft topics' agenda. This is the heart of the problem.

    What actually separates leaders from laggards

    Back to BCG: 'future-built' companies invest 120% more in AI transformation than the rest of the field. But the key difference isn't financial. The key difference is where they allocate their effort.

    Leaders:

    • anchor their AI strategy in the business vision — they don't buy AI for AI's sake; they embed it inside existing goals
    • invest in people and processes versus technology at roughly 70:30 — the opposite of the rest of the market
    • train more than half of their workforce in AI capabilities; laggards plan to train about one-fifth
    • design new roles and career paths — instead of forcing AI into existing, unchanged positions
    • actively manage trust and a culture of experimentation — because they know without it the whole investment collapses

    In other words: AI leaders are companies that understand technology adoption is a change management project — not an IT project.

    The limits of AI: what artificial intelligence won't do for you

    Time for some honesty. AI handles business strategy well — data analysis, scenario modeling, generating decision options. Csaszar et al. (2024) showed that large language models can generate and evaluate strategies at a level comparable to human experts.

    But AI won't write your mission. It won't generate a credible vision. And it certainly won't connect strategy with mission and vision in a way that's authentic and motivating to employees. This isn't a limitation of current models that will disappear with the next generation. It's a limitation in principle — because mission and vision are artifacts of human meaning, identity, and values. AI doesn't possess any of the three.

    That's why real AI transformation in an organization isn't about replacing humans. It's about a deliberate division of roles between humans and AI — with full clarity about where each side has the advantage, and where its limits begin.

    What 'psychologically informed' AI adoption looks like

    At collaboration.tech we examine AI transformation as a sociotechnical problem. McKinsey's January 2025 investment intentions and full-maturity self-assessments describe different measures, not a quantified gap in value. Concretely:

    Measure the sources of trust problems with AI. Don't assume people will 'get used to it.' Diagnosis precedes intervention. You need to know what specifically in the employee's experience with AI is generating distrust — whether it's hallucinations, opacity, or a sense of lost control.

    Separate social processes inside the team from transactions with AI. They aren't the same — even though they happen inside the same tools. Team cohesion has to be designed completely differently from collaboration with an AI agent.

    Monitor and influence work motivation. Especially where 'algorithmic management' enters the picture. In an editorial synthesizing prior research, Bowdler et al. (2026) describe how intensive algorithmic management can heighten psychosocial risks — including workload intensification, reduced decision authority, and social isolation — by shifting the balance between job demands and available resources. That's not a price worth paying for efficiency.

    Measure attitude congruence, not the average. Average team sentiment toward AI is misleading. The distribution — which half is enthusiastic and which half is afraid — determines actual dynamics.

    Design capability development to reduce skill-erosion risk. Rinta-Kahila et al. (2023) describe skill erosion in a case study of accounting automation. It concerns earlier cognitive automation, not a study of modern GenAI; transfer to another technology or workplace requires investigation.

    What you'll get by reading this blog

    In upcoming posts on collaboration.tech we will:

    • show concrete organizational-psychology tools that work in AI transformation
    • analyze real-world company cases — both successes and painful failures
    • explain the latest research at the intersection of AI, neuroscience, work psychology, and organizational design
    • give you decision frameworks — no hype, no marketing, just an evidence-based approach

    If your company is planning AI investment and examining its maturity, use evidence from your own workflows to assess the value achieved.

    In the next article we'll unpack psychological safety in the AI era — with concrete measurement and concrete interventions that work.

    Sources

    #AI adoption#AI in the workplace#AI transformation#organizational psychology#psychological safety
    Share