Quality & Coaching
QA scorecards, compliance checks, keyword alerts and coaching workflows.
Everything here works off recorded calls and their transcripts, so it requires recording to be enabled and — for the automated parts — a plan that includes AI transcription.
QA scorecards
Admin → Reporting & Insights → QA Templates defines what a good call looks like for your operation. Supervisors then score calls against that template from the QA Scorecards page.
Scores attach to the call record and roll up into agent performance, so a coaching conversation can point at the specific call rather than a general impression.
Compliance checks
Admin → Reporting & Insights → Compliance Templates defines phrases and behaviours that must — or must not — appear on a call. Recorded calls are checked against the template automatically and flagged results surface on the Compliance page.
Typical uses: confirming a required disclosure was actually read, or catching language that should never be used.
Keyword alerts
Keyword Alerts watches transcripts for phrases you define and raises an alert when one appears. Unlike compliance templates, which score a whole call, this is for catching single moments — a competitor named, a complaint escalating, a compliance phrase missed.
Coaching
Coaching is the workflow that turns scores into conversations: assign coaching to an agent against specific calls, track it through, and review progress.
Performance Digests send periodic summaries so coaching is driven by a trend rather than the last call anyone happened to listen to.
Coaching Settings under Admin controls how alerts and digests are generated.
Live supervision
From the Supervisor view, Agent Monitor shows who is on a call right now, and a supervisor can listen in silently on an active call.
Listening is silent only — the platform does not currently offer whisper coaching or barging into a live call. Coaching happens after the call, against the recording.
Why stereo recording matters here
Browser recordings capture the agent on the left channel and the contact on the right. Transcription splits those channels, so every line in the transcript is attributed to the right speaker by construction rather than by a diarisation model guessing.
That makes automated scoring meaningfully more reliable — "did the agent say the disclosure" is answerable without ambiguity about who spoke.
Something here wrong or missing? Tell us — these pages describe the platform as it actually behaves, so a mismatch is a bug we want to know about.