QA Analytics Overview
QA Analytics is the part of Omniflow that grades every conversation — voice, chat, email, social — against a scorecard you define, and rolls the results into per-agent, per-team, and per-period dashboards. Manual QA samples around 2% of calls; AI-graded QA scores 100%. The point is not to replace your coaches but to make sure they spend time on the conversations that actually need attention.
Why this exists
Quality programs have always had a sampling problem. Coaches listen to a small slice of calls, score them with a rubric, and hope the slice is representative. It usually isn’t — agents perform differently on edge cases, and edge cases are exactly the calls that don’t make it into the sample. Omniflow QA scores everything against the same rubric you’d use manually, so the conversations a coach reviews are the ones the data flagged, not the ones that happened to be picked at random on Friday afternoon.
Key concepts
- Scorecard — the rubric. A set of criteria (e.g. “empathy”, “first-call resolution”, “disclosure compliance”), each scored 1–5 with a one-line rationale.
- Custom KPI — a yes/no rule layered on top of the scorecard. “Did the agent quote disclosure X?” “Did sentiment improve by the end?”
- Run — a single conversation graded once. Re-grading produces a new run; history is preserved.
- Dashboard — the rolled-up view: per-agent, per-team, per-channel, per-period.
- Outlier alert — automation that fires when a run dips below a threshold you set.
What QA does
| Capability | What it gets you |
|---|---|
| AI-graded scorecards | Every conversation scored on each rubric criterion. |
| Custom KPIs | Yes/no rules you define on top of the rubric. |
| Trend dashboards | Per-agent, per-team, per-channel, per-period. |
| Outlier alerts | Below-threshold runs ping Slack, email, or a webhook. |
| Transcript search | Free-text search across every conversation. |
| Coaching loop | Low-scoring calls feed back into Training as remedial scenarios. |
How a score gets produced
- A conversation finishes — voice, chat, email, anything.
- The transcript is finalized and any audio is transcribed.
- The QA grader reads the transcript, the active scorecard, and any custom KPI rules.
- It returns per-criterion scores and a one-line rationale for each.
- Scores land in the QA dashboard within ~30 seconds.
The QA grader uses the same rubric vocabulary as Training. Improvements in one move the other — they’re literally the same numbers.
First 10 minutes — turning QA on
Decide where transcripts come from
If conversations already happen inside Omniflow, transcripts are already there. If you want to grade calls that happen elsewhere (e.g. legacy contact center), import them. See Importing Transcripts.
Pick or build a scorecard
Start with one of the built-in scorecards — Support, Sales, or Compliance — and edit it down to the criteria your team actually cares about. Five to seven criteria is the sweet spot. See Scorecards.
Add the must-pass rules as custom KPIs
Anything that’s binary and high-stakes — disclosures, escalation triggers, identity verification — belongs in custom KPIs, not in the 1–5 rubric. See Custom KPIs.
Wait a day, then look at the dashboard
Let the grader run on a real day’s traffic before you tune anything. Open Reports & Trends and look at the distribution per criterion.
Calibrate against a hand-graded sample
Pull 20 conversations, hand-grade them with your coaches, and compare to the AI scores. Disagreements are usually rubric ambiguity — fix the rubric, not the grader. Repeat monthly.
Wire up the coaching loop
Configure outlier alerts (Notifications) and route low-scoring calls into Training as remedial scenarios (Reviews & Coaching).
What QA isn’t
- Not a replacement for human review. Coaches still read calls. QA tells them which ones.
- Not a performance management tool. Use it to coach, not to discipline. Scores are noisy on small samples.
- Not perfect. Calibrate the rubric monthly and adjust descriptions when scores drift.
Where to go next
Related
| If you want to… | Go to |
|---|---|
| Use the same rubric in training | Scoring Rubric |
| Send low-scoring calls to a Slack channel | Notifications |
| Inspect a specific call | Activity Logs & Traces |