Skip to main content
The Quality & Testing workspace groups three ways of checking whether AI Assistant is actually helping:
  • CSAT & Outcomes
  • Knowledge Gaps
  • Simulation

​
CSAT & Outcomes

The Outcomes view uses Zendesk webhook data to connect customer satisfaction and resolution data back to Assistant work. Current metrics include:
  • CSAT Score
  • Good Ratings
  • Bad Ratings
  • Average Resolution
You can review individual ticket outcomes with ticket ID, rating/comment, resolved state, resolution time, and date. The view supports 7, 14, 30, 60, and 90-day periods.
Outcome data depends on the relevant Zendesk webhook events reaching RAUM. If the view is empty, verify the Zendesk automation and satisfaction trigger before concluding that customers have achieved perfect emotional neutrality.

​
Knowledge Gaps

Knowledge Gaps identifies recurring cases where RAUM had low confidence or lacked suitable matching knowledge. Open gaps can be used to seed Learn from past tickets and draft a new procedure from representative cases. Learn about Knowledge Gaps →

​
Simulation

Simulation runs the Assistant pipeline against a Zendesk ticket without performing real mutating actions. This is useful before releasing changes to:
  • procedures;
  • models;
  • AI Guidelines;
  • tool connections;
  • Agent Mode configuration; or
  • automation scope.

​
Tool behavior in Simulation

Simulation preserves an important distinction:
  • SOURCE tools run normally so RAUM can retrieve real data from connected systems.
  • ACTION tools are intercepted so you can see what would have executed without creating the side effect.
The full reasoning/tool/answer pipeline still runs, giving you a realistic preview without turning a test into a refund, cancellation, or other regrettably authentic business event.

​
What to inspect after a Simulation

Check:
  1. which procedure was selected;
  2. which knowledge sources were retrieved;
  3. whether SOURCE evidence was correct;
  4. which ACTION calls were proposed;
  5. the planned Zendesk changes;
  6. the final answer;
  7. confidence; and
  8. any missing information or unexpected assumptions.
1

Make one controlled change

Edit a procedure, guideline, model, or connection.
2

Run Simulation

Test several representative tickets, including edge cases.
3

Validate in Copilot

Let real agents review generated drafts without autonomous delivery.
4

Monitor outcomes

Watch CSAT, negative feedback, confidence, Knowledge Gaps, and Activity.
5

Widen automation deliberately

Expand Agent Mode confidence bands or trigger scope only when the evidence supports it.

Agent Mode

Apply confidence-based delivery after testing.

Activity

Inspect the exact run when a test or live ticket behaves unexpectedly.