Testing is essential for refining AIPs and improving automation rates. Cevro provides two ways to test your procedures before they affect live conversations.
Two Ways to Test
When editing an AIP, click the Preview button in the top right to open an embedded chat widget.
- Tests with draft versions of all AIPs
- Quick way to verify changes without leaving the editor
- Reloads when you save and click Preview again
2. Playground Page
Navigate to Workspaces → Playground for a full-page testing environment with advanced options.
The Playground
The Playground is the primary testing environment for your AIPs. It offers granular control over test scenarios.
Brand Selection
Select which brand to test against. This affects:
- Which brand variant of AIPs gets used
- Brand-specific Knowledge Base articles
- Brand-specific backoffice data
Agent Selection
Choose which AI agent persona to use for the test. Useful for testing different tones and styles.
AIP Selection
Skip classification and jump directly to a specific AIP—even inactive ones. This lets you test:
- Draft AIPs before publishing
- Inactive AIPs before activating
- Specific procedures without needing to phrase the right trigger message
Channel Toggle
Switch between Chat and Email modes to see how responses differ:
- Chat mode: Conversational, shorter responses
- Email mode: More formal, complete responses
Testing with Backoffice Data
Mock Backoffice Data
Enable Use Mock Backoffice to test without hitting your real backoffice. Describe the scenario you want to simulate in natural language:
Player has a pending withdrawal of $500 submitted 3 days ago.
VIP level is Gold.
Last deposit was $200 on December 1st.
Real Backoffice Data
Disable mock mode to use your actual backoffice. You can also:
- Pre-authenticate the player by providing auth field values (email, session ID, etc.)
- Test with real player data from your backoffice integration
Draft vs Published
AIPs have two states:
| State | Who sees it | Editable |
|---|
| Draft | Only you (in Preview/Playground) | Yes |
| Published | Live players | No (read-only) |
How It Works
- Edit in Draft mode — Make changes to instructions, settings, etc.
- Save — Click the Save button to save your draft (doesn’t affect live)
- Test — Use Preview or Playground to verify behavior
- Publish — When satisfied, click Publish to push changes live
The Preview and Playground always use draft versions. If you’re seeing unexpected behavior, check if there’s an unpublished draft that differs from the published version.
Testing Tips
Use real player messages — Test with actual phrases players use, not idealized examples
Test edge cases — Try ambiguous requests, frustrated tones, incomplete information
Test the full flow — Don’t just test that the AIP triggers—test the complete conversation path
Test brand variants — If you have brand-specific variations, test each one using the brand selector
Check escalation behavior — Verify that escalation criteria work as expected
Test with different locales — Use the locale selector to test how the AI handles different regions
Testing Player Context
If your AIP uses Player Context (backoffice data):
With mock backoffice:
- Describe the data scenario you want to simulate
- Cevro generates realistic fake responses
With real backoffice:
- Disable mock backoffice
- Enable pre-authentication
- Fill in the auth field values (email, session ID, etc. per your integration)
- Cevro will fetch real data for that player
Common Issues
| Issue | Cause | Fix |
|---|
| AIP not triggering | Description doesn’t match test message | Refine the description or example utterances, or select the AIP directly |
| Old behavior showing | Testing with published version | Ensure you’re testing drafts (Preview/Playground use drafts by default) |
| Player Context empty | Not authenticated or backoffice error | Enable pre-authentication and fill in auth fields, or use mock backoffice |
| Wrong AIP selected | Multiple AIPs match the input | Make descriptions more distinct, or select the AIP directly to skip classification |
