Scoring Rubric

A rubric is the set of criteria a call is scored against. Omniflow ships a built-in rubric that covers most use cases; you can also build custom rubrics per scenario or per team.

The built-in rubric

Criterion	What it measures	1-5 scale
Empathy	Did the agent acknowledge feelings before solving?	1 = ignored, 5 = textbook acknowledgment
Resolution	Did the customer’s problem get solved?	1 = not solved, 5 = solved cleanly
Compliance	Were required scripts, disclosures, or policies followed?	1 = missed, 5 = nailed
Efficiency	Time-to-resolution given the scenario’s complexity.	1 = wandered, 5 = direct without rushing

Each criterion is scored 1-5 by the AI grader; the overall score is a weighted average.

The built-in rubric is also what runs against production calls in QA Analytics. The same scores you tune in training are the ones you’ll see on real customer conversations.

Custom rubrics

Build a custom rubric when the built-in one doesn’t fit — e.g. a financial-services team needs disclosure_quoted_verbatim, a sales team needs next_step_committed.

Custom criteria fields

Field	What it does
Name	Short label shown on dashboards.
Description	What “good” looks like — the AI grader uses this.
Weight	0-1, must sum to 1 across the rubric.
Mandatory minimum	Optional. If the score is below this, the attempt fails regardless of overall average.
Anchors	1-5 narrative anchors so graders calibrate.

Example: sales discovery rubric

Criterion	Weight	Mandatory min
Asked a discovery question in the first 60s	0.20	3
Mirrored the customer’s stated goal	0.20	—
Quantified the customer’s problem	0.15	—
Proposed a relevant solution	0.20	3
Committed a next step	0.25	4

Committed a next step has a mandatory minimum of 4 — if the call ends without a clear next step, the attempt fails regardless of how well everything else went.

Calibration

Run the same call past the AI grader, a senior coach, and a junior coach. Compare scores. Where they diverge, refine the criterion description until the AI’s interpretation matches the senior coach’s.

Don’t lock a rubric until it’s calibrated. A miscalibrated rubric grades the wrong things, and trainees lose trust in the score quickly.

Weighting tradeoffs

Lots of low-weight criteria → noisy scores, hard to act on.
A few high-weight criteria → clearer signal, but easier to game.
Mandatory minimums → catch catastrophic failures even when the average looks fine.

Most production rubrics end up with 4-6 criteria and 1-2 mandatory minimums.

Open in Omniflow

If you want to…	Go to
Apply the rubric to production calls	Scorecards
Define custom KPIs in QA	Custom KPIs
Coach a low-scoring attempt	Reviews & Coaching