You're Measuring AI Adoption Wrong

Cover of You're Measuring AI Adoption Wrong PDF · 7 pages ↓

Most companies measure AI adoption by weekly active users. This whitepaper argues that the number shows who opened the tool and says little about value. It explains why $30Bn in enterprise AI spend has had no P&L impact, describes five levels of AI fluency, and sets out what helps people move up.

What’s inside

  • Three measurement mistakes. Confusing activity with value, counting time saved on the same output, and tracking weekly active users. One cited METR trial found experienced developers took 19% longer with AI tools while feeling 20% faster.
  • What fluency looks like. In Anthropic’s AI Fluency Index (February 2026, nearly 10,000 conversations), the top fluency behavior was iteration at 85.7%. Next came clarifying the goal (51.1%), providing examples (41.1%) and setting the interaction mode (30%).
  • The five stages of AI capability. The Casual User, the Deliberate Prompter, the GPT Builder, the Workflow Architect and the Force Multiplier. Stages 1 to 3 improve personal efficiency. Stages 4 and 5 change what the organization can do.
  • Where most companies stand. Typically 80% of the workforce is at Stage 1, with almost nobody at Stages 4 and 5. Training alone rarely changes this. The work environment does.
  • Four things that break the plateau. Solving the blank page problem, letting anyone build without code, turning individual wins into team-wide patterns, and giving leaders behavioral metrics such as iteration depth, workflow creation rate and sharing frequency.

Key takeaway

The models are good enough. The obstacle is behavioral, and the right question is whether AI is changing what your people are capable of.

Bring your agent. Bring your team.