Measuring AI Readiness: Beyond Tool Adoption

AI readiness is commonly measured by tool adoption: which models are licensed, which integrations are turned on, how many seats are issued. These metrics correlate weakly with whether the organisation produces better outcomes with AI. We describe a behaviour-based readiness assessment that scores teams on how they decompose work for agents, how they design feedback loops, and how they handle disagreement between human and model. The instrument is short, repeatable, and yields a different ranking than tool-adoption surveys.

Adoption tells you what was bought, not what is used or what works. Two teams with identical tool stacks routinely produce very different outcomes, and the difference is rarely about the tools.

• 01Decomposition: does the team factor work into units an agent can actually finish?
• 02Feedback: how short is the loop between an agent's output and a verifiable signal?
• 03Disagreement: when a human and a model disagree, what happens, and how is it recorded?
• 04Iteration: do prompts, skills, and evaluations get versioned, or are they rewritten from scratch each time?

• 01High-tool, low-readiness teams: lots of seats, no compound learning.
• 02Low-tool, high-readiness teams: a small surface area but tight loops, often outperforming their better-resourced peers.
• 03Stuck-in-pilot teams: the readiness score plateaus when feedback loops never close.

A short instrument cannot replace working with a team. The score is a starting conversation, not a verdict, and it does not currently account for domain-specific constraints (regulated industries, safety-critical work).