Autonomy readiness scores a candidate AI project not on how valuable it could be, but on whether the controls around it are built for the independence it is about to exercise. Three questions do most of the work: what systems can this agent touch once it is live, what happens the moment it takes a wrong action and how fast can a person catch or undo it, and who has to approve its output before it reaches a customer, a contract, or a ledger. A low score means the answer to one of those is “nobody has built that yet.”
It matters most as one input among several, because it behaves differently from the others. Strategic fit, data readiness, and expected ROI all describe how good an idea is. Autonomy readiness describes whether it is safe to run the way the pitch describes, which is a precondition, not a strength that can be traded off against the rest. Treating it as just another number in an average lets a project with excellent data and an obvious return win a ranking despite having no logging, no approval step, and no way to catch a bad action before it lands.
The practical fix is to let a low autonomy readiness score cap what a project is cleared to do rather than pull its total down. A project scoring poorly here doesn’t get shelved: it launches at a lower tier of independence, checks-and-reports or draft-and-approve instead of fully autonomous, and earns a higher score once the gap it exposed is closed.