The McKinsey AI Paradox: Why 78% of Companies Use AI But Few See ROI

The headline number

In McKinsey's global AI survey, 78% of organizations said they use AI in at least one business function. But only 5.5% called themselves 'AI high performers' — organizations attributing more than 5% of their EBIT (bottom-line profit) to AI. By the 2025 survey, adoption had climbed to 88% — yet the high-performer share barely moved, to about 6%. Wider adoption did not close the value gap.

Source: McKinsey — "The State of AI: How organizations are rewiring to capture value" (QuantumBlack).

The same tool, opposite outcomes

A 2025 randomized controlled trial by METR gave experienced open-source developers real coding tasks, then randomly allowed or disallowed AI tools per task. Developers with AI available were 19% slower — despite expecting to be 24% faster going in, and still believing afterward that AI had sped them up by 20%. Undisciplined, unscoped use made experienced people slower while feeling faster.

In GitHub's own controlled study, by contrast, developers using GitHub Copilot on a defined coding task finished 55% faster on average — a statistically significant result, same category of tool, opposite outcome.

Sources: METR — "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity"; GitHub Blog — "Research: quantifying GitHub Copilot's impact on developer productivity and happiness."

The difference is discipline, not the AI

The gap between these two studies isn't about which tool is better — it's about how disciplined and well-scoped the usage was. This is the same pattern behind the McKinsey adoption-vs-impact gap: most organizations (and most individuals) are using AI at the equivalent of 'basic chat and search' — one-off questions with no repeatable process — while the 5.5%-6% seeing real returns have built structured, connected, repeatable workflows around it.

The 5-rung ROI ladder

Climbing from where most people sit to where the high performers sit isn't about a better model — it's about moving up five rungs: basic chat & search, summarizing documents, structured document work (Projects, standing instructions), connected repeatable workflows (Skills, Connectors, MCP), and finally agentic work that builds and ships something end to end. Where you sit on this ladder — not which plan you're paying for — is what actually decides the outcome.