Proposed lead: Nikola Mrkšić, with the buyer’s IT and operations owners.
Why it matters: NIST’s generative-AI profile addresses incident responsibilities, monitoring and recovery. Retell’s deployment-strategist job description also explicitly includes repeatable deployment playbooks—useful evidence that operational delivery is a defined activity, not proof of any particular staffing requirement. (nvlpubs.nist.gov)
Question to put to the expert:
When a model, prompt, voice, integration or business policy changes, who must establish that the live agent still works—and who can stop it?
A strong answer should cover:
- Versioned configurations and a record of material changes.
- Regression tests tied to the buyer’s important failure modes.
- Staged release, rollback and a tested alternative call route.
- Separate alerts for technical availability and business-outcome deterioration.
- Ongoing human review, including low-volume but consequential exceptions.
- Clear ownership of alert triage, caller recovery, root-cause analysis and supplier escalation.
Evidence to request: A responsibility matrix, example change record, incident runbook and an agreed monitoring budget.
Follow-up: At 2 a.m., whose phone rings when bookings stop appearing but the agent continues telling callers they are booked?
Expert quote: [Reserved for an approved interview quotation.]