> ## Documentation Index
> Fetch the complete documentation index at: https://docs.regentra.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Automations & AI Reviews

> Automate ticket triage and assignment; review AI-suggested replies before they reach the customer.

Reduce manual work with automation rules that react to ticket events. Review AI-generated replies before they're sent so you stay in control of the customer experience.

## What it does

**Automations:**

* **Trigger events** — react to ticket creation, status changes, replies, and SLA escalations
* **Conditions** — match on priority, status, category, contact email, company, and more
* **Actions** — auto-assign, change priority, update status, add internal notes, send notifications
* **Rule priority** — execute rules in order; stop processing after a match if enabled
* **Execution logs** — view every time a rule ran, whether it matched, and what it did

**AI Reviews:**

* **Pending reviews** — see unreviewed AI-suggested replies, sorted by confidence
* **Condition matching** — understand why the AI suggested each action
* **Override ratings** — mark suggestions as Correct, Wrong, or let the AI learn from overrides
* **Execution audit** — track what AI suggestions were accepted vs. rejected by your team

## How to create an automation rule

<Steps>
  <Step title="Open Automations and click New Rule">
    In the PSA module sidebar, click **Automations**, then click **New Rule**.
  </Step>

  <Step title="Name and describe the rule">
    Enter a name (e.g. "Auto-assign Critical tickets to Tom") and optional description.
  </Step>

  <Step title="Choose a trigger event">
    Ticket Created, Status Changed, Reply Added, or SLA Escalation.
  </Step>

  <Step title="Add conditions (optional)">
    Click **+ Add Condition**. For each, choose a field (e.g. Priority), operator (Equals), and value (CRITICAL). All conditions must match for the rule to execute.
  </Step>

  <Step title="Add actions">
    Click **+ Add Action**. Choose: Assign Ticket, Change Priority, Change Status, Add Internal Note, or Send Notification. Configure the payload (e.g. which technician).
  </Step>

  <Step title="Set priority and options">
    Lower Priority = executed first. Check **Enabled** to activate. Check **Stop after match** to prevent other rules from running if this one matches.
  </Step>

  <Step title="Save the rule">
    Click **Create Rule**.
  </Step>
</Steps>

## How to test an automation rule

<Steps>
  <Step title="Open the rule">
    Find the rule in your Automations list.
  </Step>

  <Step title="Click the test icon">
    A dry-run executes the rule against a sample ticket. You'll see the test ticket evaluated, which conditions matched, what actions would have executed, and a "Would Execute" / "Would NOT execute" summary.
  </Step>

  <Step title="Review and adjust">
    If the test shows unexpected behavior, edit the conditions or actions and test again.
  </Step>
</Steps>

## How to review AI suggestions

<Steps>
  <Step title="Open AI Draft Reviews">
    In the PSA sidebar, click **AI Reviews** (admin-only entry, near the bottom of the sidebar).
  </Step>

  <Step title="View pending suggestions">
    The Pending Reviews tab shows tickets with unreviewed AI replies, sorted by confidence (low confidence first).
  </Step>

  <Step title="Open a ticket and review the AI reply">
    Click a ticket to view the suggested reply, its confidence score, and the trigger.
  </Step>

  <Step title="Accept, edit, or reject the reply">
    * **Accept** — send the reply as-is to the customer
    * **Edit** — modify and send your version. The system logs the override.
    * **Reject** — don't send the reply. The system logs the rejection.
  </Step>

  <Step title="Provide feedback (optional)">
    Add a reason for rejection or edit. This feedback trains the AI model over time.
  </Step>
</Steps>

## Understanding AI confidence

AI replies include a confidence score (0-100%) representing how sure the model is. Low scores mean the AI was uncertain — review carefully. **Low confidence does not always mean wrong; high confidence does not guarantee correct.** Always review the actual reply text and the ticket context before sending.

## AI and ticket priority

The AI auto-analyzer suggests, **never sets**, ticket priority. The priority chosen at ticket creation — by the customer (portal or Teams bot), by the inbound-email classifier, or by the operator (manual UI) — is the sole source of truth.

| Surface                                                           | Behavior                                                                                                                                              |
| ----------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
| AI's per-ticket triage analysis                                   | Stored in `aiAnalysis.suggestedPriority`, visible in the ticket sidebar with an Apply control                                                         |
| Automatic write to `ticket.priority` from the AI analyzer         | **Never happens.** No upgrade, no downgrade — the AI cannot mutate priority on its own                                                                |
| Operator-authored automation rule with **Change Priority** action | Allowed. MSP rules are explicit operator policy and remain authoritative                                                                              |
| VIP override                                                      | Bumps LOW/MEDIUM → HIGH for VIP-tagged contacts (never downgrades)                                                                                    |
| Ticket-spike pattern detection                                    | Bumps LOW/MEDIUM → HIGH when 3+ tickets arrive from the same company within 2 hours (never downgrades)                                                |
| AI escalation step (security or high-frustration concern)         | Bumps LOW/MEDIUM → HIGH AND flips status to **Escalated** with a visible audit reply explaining why. Never downgrades a customer-set CRITICAL or HIGH |

<Note>
  This is a hard guarantee. A customer-submitted CRITICAL ticket will arrive in the PSA at CRITICAL regardless of what the AI's per-ticket analysis suggests. Notification rules and SLA timers keyed on customer priority always fire correctly.
</Note>

<Info>
  The **Change Priority** automation action listed in the rule-builder above is for operator-authored MSP rules. It does not represent AI behavior. Use it freely — your rules are policy you define, not AI judgment.
</Info>

## Frequently asked questions

<AccordionGroup>
  <Accordion title="Can I test a rule without creating it?">
    No. Save the rule first, then use the test button. Drafts cannot be tested.
  </Accordion>

  <Accordion title="What happens if two rules match the same ticket?">
    Rules execute in priority order (lower number = first). If a rule has **Stop after match** enabled, execution stops there. Otherwise all matching rules execute in sequence.
  </Accordion>

  <Accordion title="Can I disable a rule without deleting it?">
    Yes. Click the play/pause icon next to the rule. It's disabled immediately and won't execute on new tickets.
  </Accordion>

  <Accordion title="Are AI reviews separate from automations?">
    Yes. AI reviews are for manual approval of AI-suggested replies. Automations execute immediately with no manual review. They can work together: an automation might trigger an AI suggestion, which then appears in the review queue.
  </Accordion>

  <Accordion title="Does the AI learn from my overrides?">
    Yes. When you mark a suggestion Correct or Wrong, or edit it, that feedback trains the model. Over time, suggestions should improve.
  </Accordion>
</AccordionGroup>