AI Operations FAQ
Answers to common setup and operating questions about AI Operations.
For what it is and how to set it up, see the
AI Operations Overview.
Setup
Do I need a licence to access AI Operations?
No. AI Operations is not a licensed app — it is available to every tenant automatically. Any user with the admin role or above sees it in the admin sidebar. It becomes useful as soon as at least one app with an AI agent domain is enabled.
I turned on an app’s AI agent but nothing is happening — what did I miss?
Two switches must be on. The app’s own agent toggle enables the agent itself, and Autonomous Operations in AI Operations (Admin → AI Operations → Agent Settings) must also be on for that domain. If that switch is off, the agent is technically present but will not propose or execute any actions.
Permissions and access
Who can approve pending actions?
Any user with the manager role or above can approve most pending actions. The exceptions are Compensation and Employee Data domains — those require admin or above to approve, because approving executes a sensitive write (pay change, employee record edit).
Can a System Admin pause agents from the Console?
Yes. A System Admin can issue a global pause from the Console’s Automation Hub. Domains paused this way cannot be resumed by a tenant admin — only the System Admin who paused them can lift the hold.
Day-to-day
What are the trust levels and how does a domain advance?
Every domain starts at Probation and can advance through three levels:
- Probation — all actions require approval. Daily cap: 10. Advances after 50 successful actions with a 95% success rate.
- Standard — routine actions auto-execute; high-impact, bulk, and critical actions still require approval. Daily cap: 50. Advances after 200 successful actions with a 98% success rate.
- Trusted — only critical actions require approval. Daily cap: 200. This is the highest level.
Advancement is evaluated daily by a background job. You can also set the trust level manually from Agent Settings — a reason is required and recorded in the audit log.
What counts as a “high-impact” action?
An action is classified as high-impact if any of the following are true: it affects more than 5 people, it involves compensation or salary, or it involves termination or offboarding.
How do I force approval for high-impact actions even at the Trusted level?
Turn on Require approval for high-impact actions in the domain’s Agent Settings panel. When this is on, actions that affect more than 5 people or involve compensation or termination always require approval — regardless of trust level.
What happens if agents accumulate too many rejections?
If a domain gets 3 or more rejected actions within 30 days, every action for that domain is routed to the approval queue regardless of its trust level. This override is additive — it can only add an approval requirement, never remove one.
How long do pending approvals last before they expire?
Pending approvals expire automatically after 14 days. Three days before expiry, the action’s owner and their manager are notified. After expiry, the action is cancelled and tenant admins are notified with the count. If the automation is still needed, you can re-run it.
Can I retry a failed action?
Yes, from the action’s detail page (Action History → click the row), use the Retry button. Retries are capped at 3 per action. Sample-data actions cannot be retried. If the action’s domain does not have a registered runner, the Retry button will not appear.
When something looks wrong
“The agent was working yesterday but now it’s doing nothing”
Check three things in order:
- Paused? Go to Admin → AI Operations → Overview. If the domain card shows a paused status, resume it (or ask your System Admin if it was paused globally from the Console).
- Daily limit hit? Check Action History for today’s count. The default cap is 10 at Probation, 50 at Standard, and 200 at Trusted. You can raise it in Agent Settings (max 500).
- Auto-paused on error rate? If the 24-hour error rate exceeds the configured threshold (default 5%), the domain pauses itself automatically. The hourly safety monitor also sends a warning when the error rate reaches 80% of the threshold, and when daily actions reach 90% of the limit. Resume the domain after investigating the errors.
The error rate check requires at least 5 actions before it activates, so a single failure will not trigger an auto-pause.
“I resumed a domain but it says only the System Admin can resume it”
That domain was paused globally from the MangoApps Console (system-admin level). Tenant admins cannot lift a console-level pause — contact your System Admin to resume it.
“An action I approved did not execute”
When you approve an action, it executes immediately. If execution fails, the action moves to the failed state. Check the action’s detail page in Action History for the error. You can retry it (up to 3 times) from there.
“Trust level isn’t advancing even though the agent is doing well”
Trust advancement counts only Completed and Failed actions — not Detected (pending), Pending approval, Approved (awaiting run), Rejected, Preview, or Duplicate skip. A standing approval backlog does not count toward the success rate. Check the domain’s stats in Agent Settings to see the current count and success rate against the advancement thresholds.
Limits and ranges
What are the configurable limits and their allowed ranges?
| Setting | Allowed range | Default |
|---|---|---|
| Max actions per day | 1–500 (0 = unlimited) | Per trust level: 10 / 50 / 200 |
| Confidence threshold | 0–100 | 95 |
| Pause on error rate (%) | 0–100 (0 = disabled) | 5 |
| Notification cooldown | 0–1,440 minutes | — |
| Approval workflow levels | 1–10 | — |
Per-trust-level overrides for max actions/day, confidence threshold, and error rate can be set in the advanced section of each domain panel.
How many rows can I export from Action History?
The CSV export covers up to 10,000 rows, ordered newest-first.
Which apps support trigger-specific approval workflows?
Three apps currently support trigger processes:
- Inspections — inspection submission, template creation, corrective action
- Frontline Execution — campaign request, campaign publication
- Policy Hub — policy publication
Other apps can have approval workflows, but without trigger-process scoping — the workflow applies to the app generically.
More help
- AI Operations Overview
- Agent Settings (Admin → AI Operations → Agent Settings) — per-domain configuration
- Trust & Safety (Admin → AI Operations → Trust & Safety) — rejection feedback and guardrail configuration
- Ask AI — the assistant answers AI Operations questions from these articles.