Score calls with AI quality assurance

Define what a good call means

Open Monitor → AI quality assurance and use QA settings to create a rubric. Configure a name, pass threshold, scoring criteria and weights. Each criterion needs a clear description of the behavior to evaluate.

For a receptionist, useful criteria include accurate clinic facts, correct escalation, verified booking confirmation, respectful handling of corrections and a clear ending. Avoid a vague criterion such as “be perfect”. Add post-call extraction fields only for data you actually need.

Evaluate a completed call

Choose Evaluate a call, select a completed session and rubric, then inspect the results. Review per-criterion reasoning and extracted information against the transcript and real action outcome. A model-generated score supports review; it is not an independent proof of a medical, legal or booking decision.

Automatic QA cohorts

Create a cohort to admit matching completed calls automatically. Configure:

• Agents, channel, status, direction and optional sentiment.
• Start/end dates in UTC and minimum/maximum duration.
• Sampling percentage and weekly admission limit.
• Maximum dispatches per minute and scan interval.
• The resolution criteria rubric.

Use recent admissions to confirm that the cohort selects the calls you intended. Saving a cohort is different from having a completed evaluation: jobs must run and the evaluation model must be available. Start with a small sample and limits you can review.

Keep rubric versions meaningful. When changing your booking policy or emergency script, update both the agent and the corresponding rubric, then inspect new evaluations for false passes and false failures.