Training Overview
Training is the part of Omniflow that turns the same AI infrastructure powering your customer-facing agents into a coaching environment for your human team. A trainee runs a voice or chat conversation against an AI-played customer, the conversation gets graded against your rubric, and a coach reviews the high-signal attempts. Because the rubric is shared with QA Analytics, what you train on is exactly what gets measured in production — improvements show up in both places.
Why this exists
Most call-center training is some combination of: shadowing a senior agent, reading SOPs, and roleplaying with another trainee. Shadowing doesn’t scale. Roleplay is awkward and inconsistent. Real customer calls are too risky for a new hire on day one. Training fills that gap with an AI customer that’s unlimited, consistent, and graded against the same rubric your real conversations are graded against — so a trainee can fail safely, see exactly why, and improve before they touch a live customer.
The four building blocks
| Block | What it is |
|---|---|
| Persona | An AI-played customer with a mood, language, voice, and backstory. |
| Scenario | A situation the trainee has to handle — objections, success criteria, supporting docs. |
| Learning path | An ordered sequence of scenarios, optionally with prerequisites. |
| Attempt | One trainee running through one scenario — recorded, scored, reviewable. |
These compose: a scenario uses one or more personas; a learning path is a sequence of scenarios; each run produces an attempt.
A typical loop
1. Manager creates a scenario → "Refund objection — gold tier"
2. Manager bundles 4 scenarios into a path → "New hire — week 1"
3. Manager assigns the path to 6 trainees
4. Each trainee runs each scenario → voice call against the AI customer
5. AI grades the attempt on a rubric → empathy, resolution, compliance
6. Manager reviews 1-2 calls per trainee → adds feedback, updates the rubric
7. Insights roll up into the team dashboardWhat’s different from a script-based simulator
- The customer is an LLM, not a branching tree. It will improvise, push back, change its mind, and sometimes go off-topic — exactly like real customers do.
- The persona has a mood, and the mood drifts during the call. A skilled trainee can de-escalate; an unskilled one will see it spiral.
- The rubric is your rubric — built-in or fully custom — and the same rubric grades production calls in QA Analytics.
- Documents matter. Attach SOPs and case docs to a scenario, and the AI customer will reference them realistically.
The training and QA modules share a rubric vocabulary. Improving training metrics moves QA metrics, and vice versa — they’re literally the same numbers.
First 10 minutes — the setup order
Building these blocks in the wrong order is the most common mistake teams make. The order matters because each block depends on the one above it.
Build a persona first
Decide who your trainee will be talking to. Start with one — “Frustrated mid-tier customer, English, calling about a delayed refund” — and refine later. See Personas.
Write your first scenario
Pick a real situation your team handles. Write the success criteria explicitly: “trainee acknowledges the delay, offers the SLA-compliant resolution, ends with a satisfaction check”. Attach the SOP or policy doc. See Scenarios.
Pick the rubric
Use a built-in rubric to start, then customize. Five to seven criteria is the sweet spot — more than that and trainees can’t track what to improve. See Scoring Rubric.
Run a practice call yourself
Before assigning to anyone else, run the scenario yourself. Read the trace. Adjust the persona’s mood, the scenario’s success criteria, or the rubric until the experience matches what you’d expect from a real call. See Practice Calls.
Bundle scenarios into a learning path
Once you have 3–5 scenarios that work, bundle them into a learning path with a clear ordering: easy → moderate → hard. See Learning Paths.
Assign and review
Assign the path to a small cohort. Don’t try to review every attempt — let AI Insights surface the ones that matter, and use Reviews & Coaching to leave feedback on those.
Close the loop with QA
After a few weeks, compare training scores to QA scores on real calls for the same trainees. The gap is your coaching opportunity.
Where to go next
Related
| If you want to… | Go to |
|---|---|
| Build voice agents that handle the same scenarios | Agent Studio Overview |
| Score real customer conversations | QA Analytics Overview |
| Send low-scoring real calls back into training | Reviews & Coaching |