Leading AI at Work: Governance, Rollout and Practice / Unit 9 LAI 207 review, 2 to 3 minutes
Capstone Three: An AI Feature Through the Gate
The idea
The capstone's third part, constructed for this course, takes the utility's payment-risk model through the book's checklists and the AI feature gate.
In practice
The model scores each overdue account for the risk that it will not pay. Its builder wants the score to set the order in which disconnections are processed. Priya runs the checklist for introducing an AI feature. The purpose is to help officers prioritize their calls; it is not for deciding whether a household is disconnected. The model reads the Account, the Payment History and the Payment Arrangement; and, through the processing order, it influences the Service Disconnection, a Gold thing. At the gate, she fills the five items, and the boundary test gives a clear answer: if nobody looks, a high score moves a household to the front of the disconnection queue, which is a decision. So the score is moved: it now orders the officers' call list, and a disconnection still needs an officer's review of the protected-status field. The safety checklist finds two gaps. The Disconnection's Pack did not yet say which fields the model may read, and there were no tests for the model's behavior on households with a medical exemption. Both are fixed before release. The rollback is a switch back to the old call order, owned by the collections manager.
Where people slip
The model was not changed at all; its place in the process was. That is the pattern the book returns to in every domain: AI may draft or prioritize around a consequential decision, but it does not make the decision. The gate found the problem while the fix was a configuration change, not an incident.
Your place on the trail
Every lesson rests on earlier ones and opens later ones. Tap any to jump.
Builds on
Something wrong in this lesson? Flag it · Sources for Leading AI at Work: Governance, Rollout and Practice