Adoption is at 64% of the workforce, up ten percentage points year on year, and daily use has nearly doubled (PwC, 2026). The rollout is working as a rollout. What has not appeared are the returns that justified it. McKinsey’s State of Organizations survey of more than 10,000 leaders finds that 81% of organisations experimenting with AI do not report any meaningful bottom-line gains, and 88% are now experimenting (McKinsey & Company, 2026a).
In Brief
- Two-thirds of workers now use AI, but 14% are pulling ahead while 56% form the engine room with limited access.
- Learning and development (L&D) access across the full workforce fell from 59% to 51% year-on-year, contracting when it needed to expand.
- Most organisations are experimenting with AI; 88% have started but 81% report no meaningful bottom-line gains.
- The ROI failure and the distribution failure are the same structural problem, caused by how the operating model was designed.
- Organisations that achieved AI returns made different structural choices; operating model redesign was the deciding factor.
The AI ROI gap is an operating model design problem. It is not resolved by investing more in the same categories.
Eighty per cent of people using AI report it improved their own productivity (Tinkoff et al., 2026). The tools are working for the people who have them. But individual productivity gains do not aggregate into enterprise returns on their own, and in most organisations, they are not doing so. The investment decision was made without the structural decisions about how AI capability would reach the workforce doing operational work.
Returns follow where access goes
PwC’s survey of nearly 50,000 workers identifies four distinct workforce groups. Fourteen per cent are front-runners: workers with high-frequency AI use, rising confidence, and strong access to learning and development. Fifty-six per cent form the engine room: workers who do the core operational work of the organisation but are not far along the AI learning curve and have significantly less access to both AI tools and the development resources that would change that (PwC, 2026). Front-runners have access to learning and development resources at nearly 80%. Engine room workers come in at under 40%.
Learning and development (L&D) access across the full workforce fell from 59% to 51% year on year, contracting when the workforce needed it to grow (PwC, 2026). One in six organisations surveyed by McKinsey has no clear C-level owner for AI adoption (McKinsey & Company, 2026a). Trust in management among engine room workers fell seven percentage points year on year.
AI tools have reached most of the workforce, but capability has concentrated in the front-runner segment. The front-runners are pulling ahead. The engine room, where 56% of the workforce does the work the organisation depends on, is falling behind.
From inside the investment, the distribution problem is invisible. Adoption metrics rise, which registers as uptake. Positive productivity reports flow from the front-runners, which registers as impact. The engine room is using tools at lower rates, which reads as a training gap that more training will close. What that reading misses is the question of where the productivity gains are sitting in the operating model: whether the 80% of AI users reporting personal productivity gains (Tinkoff et al., 2026) are distributed across the workforce or concentrated in the segment already furthest ahead.
Most organisations invested in AI tools and training programs. Most did not make the corresponding structural decisions: how AI tools and development pathways would reach the engine room differently from the front-runners, how accountability for AI capability in operational roles would be located in the operating model, and who would be responsible for that trajectory. BCG’s research on frontline workers finds that 66% receive limited or no guidance on how to use the time that AI creates (Beauchene et al., 2026). The tools reached most of the organisation; the structural decisions about how access translates into capability did not.
What 81% have in common
McKinsey’s State of AI survey of nearly 1,720 participants reports that 80% of AI users see individual productivity gains, but only 37% attribute any earnings impact to AI, a figure unchanged from the previous year even as enterprise-scale AI adoption rose from 38% to 44% (Tinkoff et al., 2026). More organisations are deploying at scale; the earnings signal has not moved. Those two figures describe a structural constraint.
Approximately 50% of companies now generate value from AI, up from around 5% in 2025; structural redesign, rather than better tools or higher adoption, drove that movement (Apotheker et al., 2026). McKinsey’s analysis finds that AI high performers, approximately 6% of organisations, were nearly three times as likely to have redesigned their workflows from the ground up rather than overlaying AI onto existing processes (Tinkoff et al., 2026). BCG’s workforce transformation research finds that future-built organisations upskill more than 50% of their employees against 20% at laggards, and are four times as likely to run structured AI-learning programs that reach across the organisation (Bedard & Beauchene, 2026).
The 50% generating value and the 81% not reporting gains both describe the consequences of two decisions: the investment decision and the structural design decision. The organisations in the first group made both.
With each cycle, the gap becomes harder to reverse. Front-runners extend their lead as the engine room falls further behind the level of capability at which returns convert. Training investment arrives at a workforce that has less access to development resources than it had the year before. The gap between individual AI productivity and enterprise returns widens, because the 56% doing the operational work the enterprise depends on has no clear development pathway toward the level at which returns convert.
The decision that produces returns
The default response to an AI return gap is to invest more in the same categories: more tools, more training programs, more change management campaigns. BCG’s research on frontline workers locates the constraint elsewhere: whether the people receiving that investment have clear direction on how to use the time and capability that AI creates (Beauchene et al., 2026). Without that direction built into how roles and teams are structured, the investment produces adoption without returns.
The organisations that redesigned workflows decided where AI capability would sit in the operating model, which roles would integrate AI into how the work is done, and who would be accountable for that integration reaching the workforce doing operational work. BCG finds that AI value comes 10% from algorithms, 20% from technology, and 70% from how people and work are designed around both (Bedard & Beauchene, 2026).
The investment decision has been made across most organisations. The structural decision about how AI access and development pathways reach the engine room remains unmade. Zen Ex Machina works with organisations where the investment decision has been made and the structural decision has not.
References
- Apotheker, J., Beauchene, V., de Bellefonds, N., et al. (2026, September 30). The formula for agentic AI value (Applied AI Index 2026). Boston Consulting Group. https://www.bcg.com/publications/2026/the-formula-for-agentic-ai-value
- Beauchene, V., Duranton, S., Martin, D., Lyon, V., & Walters, J. (2026, June 3). AI at work: Why strategy matters more than tools. Boston Consulting Group. https://www.bcg.com/publications/2026/ai-at-work-why-strategy-matters-more-than-tools
- Bedard, J., & Beauchene, V. (2026, February 4). AI transformation is a workforce transformation. Boston Consulting Group. https://www.bcg.com/publications/2026/ai-transformation-is-a-workforce-transformation
- McKinsey & Company. (2026, February 19). The state of organizations 2026: Three tectonic forces that are reshaping organizations. McKinsey & Company. https://www.mckinsey.com/capabilities/people-and-organization/our-insights/the-state-of-organizations
- PwC. (2026, September 29). Global workforce hopes and fears survey 2026: Unlock the workforce advantage. PwC. https://www.pwc.com/gx/en/1/issues/workforce/hopes-and-fears.html
- Tinkoff, D., Van der Veken, L., Chui, M., & Balakrishnan, T. (2026, August). The state of AI in 2026: On the road to ROI. McKinsey & Company. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
