Voice Agent Containment Audit (QA)

Voice Agent Containment Audit (QA)

1. 01

   Pull a 14-day sample of at least 200 resolved conversations per client queue

   Stratify the sample by channel (voice, SMS, chat) and by intent so a single high-volume queue does not dominate the review.

2. 02

   Replay each AI-resolved call against the client's written resolution criteria

   Score containment as correct only when the outcome matches what a trained human agent would have done, not merely when the call ended without a transfer.

3. 03

   Flag every warm transfer and measure the handoff reason distribution

   Platforms such as Nectar Desk route unresolved calls to live agents, so the transfer log is the fastest read on where the bot's knowledge stops.

4. 04

   Compare containment rate against the vendor's published range and the client's baseline

   A 40-65% autonomous resolution claim is a category benchmark, not a guarantee; report the gap in points and the dollar value of the delta.

5. 05

   Insert a human review checkpoint before any agent-generated summary reaches the client dashboard

   Fewer than half of agent-generated insights lead to meaningful action, so unreviewed summaries inflate perceived performance and erode trust at renewal.

6. 06

   Test 20 adversarial and edge-case utterances per queue

   Include accented speech, background noise, overlapping speakers, and out-of-scope requests; log misroutes and silent failures separately.

7. 07

   Document the retraining actions and owners in the client-facing QA memo

   Name the intent clusters to retrain, the expected containment lift, and the date of the next audit so the retainer review has a measurable commitment.