Skip to main content

​
AI auto-QA readiness card

The Quality hub in the dashboard carries one card that answers “is the automated AI scoring pipeline actually covering my calls” before you trust its scorecards. This guide reads that card end to end: what each block measures, where the next-step chips point when a gate is missing, and how to query the same snapshot programmatically. For how the pipeline itself works — the toggle, the flag threshold, the sweep that writes the scorecards — see AI Auto-QA configuration. This guide is about the readiness surface that tells you it is set up correctly.

​
1. How the panel reads

Open Quality in the dashboard. The readiness card (“AI auto-QA scorecard coverage”) renders four blocks, top to bottom.

​
Generation gates

Two pre-flight gates decide whether the sweep scores any call at all: When both gates pass, the card shows a green Pipeline active badge and no chips. When a gate fails, the card renders one chip per failed gate — “Turn on AI auto-QA” or “Activate a scorecard form” — each linking straight to the surface that fixes it. Zero chips is the healthy steady state; the card names the missing gate rather than leaving you to guess.

​
Coverage gauge

Below the gates, a Coverage bar reports the share of eligible calls — calls attributed to an agent, with a transcript — over the trailing window (three days) that carry an AI scorecard. A caption under the bar counts the scored calls against the pending ones. Coverage is the denominator check for “nothing is being scored”: if the bar is low while the gates are green, the sweep’s eligibility rules are excluding traffic — calls without transcripts, or calls with no assigned agent, never reach the scorer.

​
AI scorecard quality

Once scorecards exist, two more bars appear:
  • Avg AI score — the mean total across AI-generated scorecards in the window. Color follows the suite-wide thresholds (green from 80, amber from 60, red below); a null reading renders muted rather than flashing green on an empty snapshot.
  • Flagged for review — the share of AI scorecards stamped below your flag threshold, waiting on a human. This bar is inverted on purpose: a rising flag share is worse, so it never renders on a success tone.

​
Empty state, not a fake zero

coverage_pct is null when the window has no eligible calls. The card renders an explicit empty state — “Coverage counts calls attributed to an agent with a transcript” — instead of a 0% bar, so a null read explains the eligibility rule rather than masquerading as total failure.

​
2. Access

The readiness card sits on the Quality hub’s card grid alongside the QA score trend card, and it follows the same scope as the rest of the suite: owners, admins, and supervisors see the full snapshot. A non-reviewer role sees a supervisor-only notice in place of the data — the same scope guard the Quality hub applies to every rollup. Because the pipeline’s knobs live under Settings and Evaluation forms, the fix chips route differently by role: a supervisor can read the gates, while an owner or admin flips the toggle and activates the form.

​
3. Programmatic equivalent

The card renders one endpoint. Query it directly when you want to gate an automation, alert on coverage, or wire the pre-flight into a deployment check:
The response reads as three blocks: A gate that reads enabled: false or has_active_form: false explains why coverage is stalled — fix the gate before reading coverage numbers. The full field-level matrix lives in the Quality Management API reference and the AI Auto-QA configuration guide.

​
4. Common failed gates

Toggle on, but no active form. The card raises the “Activate a scorecard form” chip, which routes to Quality → Evaluations. Author a form there (or over POST /api/v1/quality/evaluation-forms), or re-enable one you retired. The sweep always scores against your most recently updated active form, so retire the ones you don’t want driving auto-score. Gates green, coverage stuck near zero. The sweep is running but eligibility is excluding traffic: only calls from the trailing window with both a transcript and an assigned agent count. Pure-IVR or bot-only calls have no agent to score and are skipped by design. Confirm transcripts are being extracted for your completed calls, then re-check the gauge. Toggle flipped, card still shows the chip. The sweep ticks once a minute and caches tenant config for up to five minutes, so a flip can take that long to propagate to the card. Wait one cache window and refresh before assuming a failed write.