Proposed lead: Brad Cleveland, with a quality lead and finance owner.
Why it matters: ICMI’s metrics discussion argues for connecting operational measures to customer and business outcomes. For this desk, the procurement implication is to define completion independently of whether a call stayed away from a person. (icmi.com)
Question to put to the expert:
How should a buyer define AI call center cost per resolution when the agent can end a conversation without completing the underlying task?
A strong answer should cover these proposed classifications:
| Classification | Evidence the buyer should require |
|---|---|
| Autonomous completion | Correct downstream state, policy checks passed and no human completion work |
| Human-assisted completion | The task succeeded, but human rescue is recorded and costed |
| Appropriate escalation or refusal | Correct handling of a boundary; reported separately from completed bookings |
| Failure | Wrong, missing or unauthorised result; unresolved abandonment; or an unfulfilled commitment |
| Unknown | Insufficient evidence to determine the result; never silently converted into success |
Define a unique task using the underlying issue or booking—not just a telephone number. Agree an appropriate follow-up window for detecting repeat contact and corrections.
Evidence to request: A disposition dictionary and a privacy-controlled reconciliation between calls, transactions and subsequent rework.
Follow-up: Can finance reproduce the completion rate without accepting the supplier’s own success label?
Expert quote: [Reserved for an approved interview quotation.]