AI Insights

Every practice attempt generates a short AI-written insight after scoring. Across the team, those insights roll up into pattern detection — the things that keep coming up, the trainees who are improving, the scenarios that are tripping people up.

Per-call insights

After an attempt is scored, Omniflow generates a 2-3 paragraph note covering:

Strengths — concrete moments the trainee did well (with timestamp links).
Improvement areas — specific turns to revisit.
Suggested coaching prompt — a one-line question a coach can use as the opener for a 1:1.
Recommended next scenario — based on what the trainee struggled with.

Sample:

Maya opened with strong empathy (“I hear how frustrating this is”) at 0:14, but missed the policy quote required for compliance — at 1:42 she paraphrased instead of reading the disclosure. Suggested coaching prompt: “Walk me through how you decide when to quote policy verbatim.” Recommended next scenario: Disclosure compliance drill.

Team-level patterns

The Insights dashboard surfaces patterns across the team:

Pattern	What it means
Repeat misses	A criterion that 30%+ of trainees missed this week.
Stalled trainees	People whose median score hasn’t moved in 4 weeks.
Improving trainees	People whose median score moved up by a configurable threshold.
Hot scenarios	Scenarios with the lowest pass rate.
Cold scenarios	Scenarios nobody runs — usually because they’re outdated.

Patterns are recomputed daily. The dashboard shows the current state and a 4-week trend.

Using insights

In 1:1s — open the trainee’s profile and use the suggested coaching prompt as the conversation opener. It saves you 15 minutes of prep.
In sprint planning — look at hot scenarios and decide whether to rewrite the scenario or build a remedial path.
In hiring — the same rubric you use to coach existing reps can be used to score role-play interviews.
In product feedback — repeat misses on compliance often surface ambiguity in the policy itself, not the trainee’s understanding.

What insights don’t do

They don’t replace coaching. The insight is a draft; the coach’s read is the final word.
They don’t grade attendance. Stalled means scores haven’t moved — not that the trainee is slacking. Always read before acting.
They don’t predict performance. Use them to find what to work on, not to rank people.

Be careful with “stalled” lists. Some plateaus are real (the trainee has reached their level), some are coaching gaps, some are personal. Talk to people before drawing conclusions.

How insights are generated

Insights are generated by Claude using the transcript, the rubric, the AI grader’s reasoning, and the coach’s review (if any). The model is configurable in AI Infrastructure; insights pricing is rolled into the per-attempt cost.

Open in Omniflow

Open Training Insights

If you want to…	Go to
Read team-wide patterns	QA Reports & Trends
Build a remedial scenario	Scenarios
Customize the rubric	Scoring Rubric