Supervisor guide: the agent scorecard drill-down
Quality → Agents → <agent> is the supervisor’s deep-link into one agent’s outcomes: QA score, adherence, AHT, FCR, CSAT, and compliance flags in one view, with the LLM-judged rubric breakdown and the drill-down to the source interactions beneath it. This guide covers how to reach the page, who can read it, what each panel means, how it contrasts with the agent’s own view, and the coaching loop it feeds.1. Reaching the page
Three paths land on the same scorecard:- Quality → Agents — the agents list links each row to that agent’s scorecard.
- Quality → Leaderboard — when a rank needs explanation, the board’s per-agent drill opens the scorecard for that row (Quality leaderboard).
- Quality → Recordings — the agent chip on any recording opens the same page, so a recording you are listening to is one click away from the agent’s full picture.
2. Who can read it — and what other roles see
- Owner, admin, and supervisor can read any agent’s scorecard. The check is enforced server-side; the UI never offers what the API would refuse.
- Every other role gets a 403, and the page answers it with a friendly “Scorecard is private” empty state rather than a generic error toast — an out-of-scope reader sees that the page exists but is not theirs to open.
- Agents never see each other’s cards. An agent’s own view is My scorecard (/me/scorecard); the contrast is in section 4.
3. What the page shows
The performance strip carries six KPI tiles over a 7 / 30 / 90-day window (default 30). Every tile is computed server-side for the selected window:
A metric the tenant does not capture yet renders as a dash, never an error.
The coaching card turns the numbers into next steps: the weakest rubrics named as focus areas, flagged interactions linked to their source, and the agent’s active coaching plan when one exists.
The rubric breakdown is the LLM-judged view: overall pass rate with the scored count and average confidence, per-rubric rows sorted weakest-first, the top failure modes, and a daily trend. Treat a low-confidence rubric row as a spot-check prompt, not a settled verdict.
Drill-down to source interactions: each evaluated interaction opens the source call on Voice → Calls; rubric rows continue to Quality → Evaluations filtered to the agent; the recordings behind those evaluations live under Quality → Recordings.
4. This page vs the agent’s own scorecard
The rubric data agrees on both surfaces; only the framing differs. When an agent asks what you see, point them to Read your own QM scorecard. The full per-panel reference for this page is Agent scorecard.
5. The coaching loop
- Open the scorecard from a leaderboard rank, a low KPI tile, or a compliance flag.
- Identify the failed rubric — the top row of the rubric breakdown is the weakest. Compare 7 days against 90 to separate a trend from a blip, and open the underlying evaluation before you conclude anything.
- Assign a roleplay in Practice — under Quality → Practice (
/quality/practice), assign the failed rubric as a roleplay so the agent rehearses it against an AI sparring partner before the next live conversation (Practice Studio roleplay). - Verify improvement in Quality → Evaluations (
/quality/evaluations): new graded conversations land in the ledger and the scorecard re-aggregates on read, so the following window shows whether the rubric recovered (QA Evaluations).
See also
- My scorecard reference — the agent self-view
- Read your own QM scorecard — the agent-side walkthrough
- Agent scorecard — per-panel reference for this page
- Quality leaderboard — the ranking that drills into this page