Readiness is a system, not a survey

The useful question is not whether a company is ready for AI. It is which decisions are ready to change without creating a new operational risk.

Most readiness work ends in a score. The score is comforting and rarely operational. A company can score well on talent and tools and still have no workflow where a model is allowed to act, no owner for a wrong answer, and no way to notice that the answer changed after a prompt edit.

We treat readiness as a property of a system, not a mood of the organization. For a named use case, five things have to be true at once: the decision is worth changing, the source material can be trusted, a person remains accountable, quality can be evaluated, and the failure mode is acceptable. If one of those is missing, the right recommendation is often to wait, or to narrow the job until the five line up.

That framing changes the first engagement. Instead of a catalogue of possible copilots, the work is to pick two or three candidate workflows and pressure-test them. What would the operator do differently on Tuesday? What evidence would make a leader stop the rollout? Which system of record has to be touched, and who will approve that?

Tools still matter. They matter after the system is named. A model, a vendor, or a platform is then chosen against a task, a latency budget, a data boundary, and a quality bar. The roadmap that follows is short on purpose. It funds the work that can be evaluated, and it refuses the work that cannot.