Scorecards

A scorecard is the AI grader’s verdict on one conversation. Each one has per-criterion scores, an overall score, a one-line rationale per criterion, and timestamped evidence quotes from the transcript.

What a scorecard looks like

Criterion	Score	Rationale
Empathy	4/5	Acknowledged feelings at 0:14 (“I hear how frustrating this is”). Missed a second opportunity at 1:42.
Resolution	5/5	Full refund issued, customer confirmed satisfaction.
Compliance	3/5	Disclosure paraphrased rather than read verbatim.
Efficiency	4/5	Solved in 3:12, slight detour into upsell.
Overall	4.0/5	—

Click any criterion to expand the evidence — the transcript turns the grader cited and links into the audio at the right timestamp.

Filtering and triage

The scorecard list supports the filters you’d expect plus a few that matter for QA work:

Below threshold — show only calls below a score floor.
Outliers — calls 2+ standard deviations off the team mean.
Unreviewed — no human eyes on it yet.
Disputed — agent disagreed with the AI score.
By criterion — find every call that scored low on compliance this week.

The default landing view is Below threshold + Unreviewed — i.e. the calls that need attention.

Override an AI score

Coaches can override any score. The AI’s number stays visible (greyed out) so the disagreement is auditable.

Open the scorecard.
Click the criterion you want to adjust.
Set your score and add a one-line reason.
Save.

The agent gets a notification. The override flows into the dashboards immediately.

Agent disputes

Agents can dispute their own scores from their dashboard. Disputes show up in the coach’s queue with the original score, the agent’s note, and a one-click approve/reject.

Outcome	What happens
Coach agrees	Score is updated, dashboards recompute.
Coach disagrees	Original score stands; coach’s note is saved with the dispute.
Coach escalates	Routed to a senior reviewer.

Take disputes seriously. A pattern of disputes on the same criterion usually means the rubric description is unclear, not that the agent is wrong.

Calibration

Once a quarter, take a sample of 30-50 graded calls and have a senior reviewer score them blind. Compare scores. If the AI is consistently 0.5+ points off on a criterion, refine the description in the rubric and re-score the period.

Exporting scorecards

Destination	How
CSV	Bulk export from the scorecard list.
Google Sheets	One-click via the Sheets integration.
Data warehouse	Stream via the API or scheduled CSV.
Slack	Daily/weekly digests via notifications.

Open in Omniflow

If you want to…	Go to
Define a new rubric	Scoring Rubric
Add a yes/no rule	Custom KPIs
Read team-level trends	Reports & Trends