How can we help?

Articles in this section

See more

All About ServiceAI's Triage Feature

Ricky Cecchini
  • Updated

ServiceAI's Triage feature uses AI to automatically categorize, route, and enhance incoming tickets from your PSA system. Instead of relying on end users to correctly classify their issues or dispatchers to manually sort through every ticket, Triage reads the subject and description of each ticket and makes intelligent decisions about how it should be handled.

This article covers everything you need to know about enabling, configuring, and monitoring the Triage feature in ServiceAI.

Availability

Triage is currently available for the following PSA/ticketing systems:

  • ConnectWise — uses separate Routing, Categorization, and Assignment rule sets
  • Autotask — uses Unified Triage Rules
  • Zendesk — uses Unified Triage Rules, with support for ticket forms and auto-discovered custom fields
  • Halo — uses Unified Triage Rules
  • Syncro — uses Unified Triage Rules
  • Kaseya BMS — uses Unified Triage Rules

Halo, Syncro, and Kaseya BMS use a manual webhook configuration in your PSA — see Troubleshooting Triage Webhook Configuration for setup details.

Usage Limits:

  • Professional Tier: 200 ticket triages per month
  • Enterprise Tier: Unlimited triages

Your usage resets monthly, and you can view your current usage at the top of the Triage settings page.

Enabling Triage

To enable the Triage feature:

  1. Navigate to Settings from the gear icon in the top navigation bar
  2. Select the Triage (beta) tab from the settings menu
  3. Toggle Enable Triage to activate the feature
  4. Click Test Permissions to verify your PSA API user has the required permissions for the Triage Server
    • If you are not able to pass the permissions check, ensure that you have entered the appropriate permissions into ServiceAI based on your specific PSA.
  5. Click Save Settings to confirm your changes

Once enabled, the full configuration options will become available, and the Triage tab will appear in your left-hand sidebar.

Testing Mode

Before deploying Triage to actively modify tickets in your PSA, it's strongly recommended to enable Testing Mode. When Testing Mode is active, ServiceAI will analyze tickets and log what changes it would have made without actually modifying anything in your PSA system.

This allows you to:

  • Observe how the AI interprets and categorizes tickets
  • Fine-tune your rules before going live
  • Build confidence in the system's decision-making
  • Identify any adjustments needed to your Rule Sets

You'll see a clear indicator on triaged tickets showing "Testing Mode - No Changes Applied" when this mode is enabled.

Spam handling in Testing Mode: When Triage detects a ticket as spam while in Testing Mode, the triage result will display "Testing Mode - Spam Detected" along with the spam confidence score. Since Testing Mode does not modify your PSA, the ticket will not be moved to your Spam Destination board — the spam detection result is recorded for your review only.

Configuring Triage Rules

Triage uses plain-language prompts to define how tickets should be handled. This approach is designed to be more intuitive than traditional rule-based systems. You can add rules by clicking the + Add Rule button next to each rule category.

Unified Triage Rules (Autotask, Zendesk, Halo, Syncro, Kaseya BMS)

For Autotask, Zendesk, Halo, Syncro, and Kaseya BMS, ServiceAI uses a single Unified Triage Rules list rather than separate Routing, Categorization, and Assignment sets. One rule list drives every decision the AI makes about a ticket, which is then evaluated in a single combined AI analysis per ticket.

This approach has two benefits:

  • Rules read more naturally because they can describe routing, categorization, and assignment in the same sentence
  • Triage runs one AI call per ticket instead of three, reducing latency and AI cost

Unified rules use each PSA's native field terminology in the prompt, so the AI is aware of the exact field names it is being asked to update.

Example unified rules:

  • "Printer and copier tickets should be routed to the Hardware Support queue and categorized as 'Hardware' with sub-issue 'Printer/Copier'"
  • "Network tickets should be assigned to Samuel Smith and use the Network Team queue"
  • "Password reset requests should use the Tier 1 Support queue and be categorized as 'Account Access'"

For Zendesk, unified rules can also reference ticket forms and their form-specific custom fields. The AI will select the appropriate form and populate its fields during the same triage pass.

Add rules by clicking + Add Rule on the Unified Triage Rules list in Triage settings.

ConnectWise Rules

ConnectWise continues to use three separate rule sets — Routing, Categorization, and Assignment — each configured independently in Triage settings.

Routing Rules determine where tickets should be sent based on their nature. Use these to direct tickets to specific boards in ConnectWise. Example: "Printer and copier tickets should be routed to the Hardware Support board."

Categorization Rules help the AI assign the correct Type and SubType. The categories must exist in ConnectWise on their respective boards. Example: "Printer tickets must have the category 'Hardware' and subcategory 'Printer/Copier'."

Assignment Rules help assign tickets to the appropriate technicians. Example: "Network tickets should always be assigned to Samuel Smith." These work best on specifically-named resources within ConnectWise.

Agent Mode Rules in Triage Briefings

The Agent Mode Rules you configure under Settings > Rules (the same Agent Mode Rules that shape AI Chat in Agent mode) now also shape the wording of the Fast Feedback agent summary internal note that Triage posts back to your PSA. This lets a single set of coaching rules govern both the in-app AI Chat in Agent mode and the post-triage briefing that lands on the assigned technician's ticket.

For example, an Agent Mode Rule like "Always confirm device patch status before recommending a reboot" will now bias the agent summary section of the triage internal note in the same way it biases AI Chat in Agent mode. Routing, categorization, assignment, and field decisions are governed by the Triage Rules above — Agent Mode Rules are scoped to the agent-summary text only and do not influence routing decisions.

If your Tenant has no Agent Mode Rules of its own, the system defaults are used instead.

Best Practices for Rules

  • Keep rules clear and straightforward to avoid contradictions
  • While there's no hard limit on the number of rules, be reasonable — too many overlapping rules can cause confusion
  • Use specific language and exact field names in quotes when possible
  • Start with a few key rules and expand as you observe the AI's behavior

Fast Feedback

Fast Feedback adds two AI-drafted outputs to every triaged ticket: an agent summary posted as an internal note for the assigned technician, and an optional auto-response posted publicly to the requester. Both are produced by the same dynamic-query workflow that powers AI Chat and the public chat experience, so a customer asking through a ticket gets the same caliber of answer as one asking through your chat page. Available for all supported PSAs.

Enable Agent Summary

When enabled, ServiceAI posts an internal note on each triaged ticket containing a concise, technician-facing summary of the issue, what's been tried, and a suggested next step. The summary now uses the full AI Chat toolset — it can mine similar tickets and KB articles while drafting — so the briefing is grounded in your tenant's own historical context.

The internal note also includes:

  • A Confidence indicator showing how confident the AI was in the underlying analysis
  • A Related tickets section listing similar past tickets, with PSA-native links so technicians can jump to the source ticket inside their PSA
  • A Related KB articles section listing the most relevant knowledge base articles, with links that resolve correctly for both external sources (ITGlue, ConnectWise KB, etc.) and articles created inside ServiceAI

Enable Auto-Response

When enabled, ServiceAI can post a public-facing reply to the requester only when the AI's confidence in the response meets your configured threshold. The auto-response runs through the same dynamic-query workflow as the public chat at /chat?tenant=…, so it can search your KB and (when allowed) consult ticket history while drafting.

Below-threshold drafts are not posted publicly — instead, they are surfaced as a dedicated internal note so you can review what the AI would have said and tune the threshold accordingly.

Every public auto-response ends with a clear AI disclaimer indicating that the reply was AI-generated, that a technician is still reviewing the ticket, and inviting feedback on response quality.

Auto-Response Template

Use the Auto-Response Template textarea below the response controls to customize the wrapper applied around the AI-drafted body. The template is a free-text field that uses {{ai-response}} as the placeholder for the AI body. Common uses include adding your MSP's signature, branding the disclaimer, or adjusting the tone of the wrapper to match your customer-facing voice.

The textarea pre-fills with the built-in default so admins can see (and edit) the operative wrapper text. Click Restore default at any time to revert to the built-in default. If your saved template equals the current default exactly, ServiceAI clears your override so future improvements to the default automatically apply.

The Auto-Response Template is only editable when Enable Auto-Response is on.

Allow Ticket Data to Be Used to Generate Response

When this toggle is on (default), the auto-response can consult your tenant's ticket history while drafting — useful when a public answer would benefit from prior resolutions for the same issue. When off, the auto-response is restricted to KB and web search, matching the toolset available to anonymous public chat.

Confidence Threshold

The Confidence Threshold controls how confident the AI must be before its drafted response is posted publicly. Default is 90%. Accepts any value from 0% to 100%:

  • 0% — every draft is posted publicly
  • 90% (default) — recommended starting point
  • 100% — only the very highest-confidence drafts post publicly

The Confidence Threshold input is enabled only when Enable Auto-Response is on. The confidence score the AI returns is its certainty in the verdict it picked (whether the drafted reply actually solves the ticket), not a probability of any single outcome — so a value of 95% means the AI is 95% certain the drafted answer is appropriate to send, in either direction.

Reviewing Below-Threshold Drafts

When auto-response is enabled but a draft falls below the configured threshold, the draft is posted as an internal note that includes the assessed confidence and the threshold in effect. This is the recommended way to tune your threshold — review a few weeks of below-threshold drafts and lower the threshold once you're comfortable with the quality at that confidence level.

Testing Mode and Fast Feedback

While Testing Mode is on, neither the agent summary nor any public response is posted to the PSA, regardless of the Fast Feedback toggles. The drafted content is still recorded in the Triage Detail view so you can preview it.

Additional Configuration Options

Monitored Boards/Queues

Specify which boards or queues ServiceAI should monitor for triage. Enter board or queue names separated by commas (e.g., "Service Desk, Help Desk, IT Support"). This allows you to control exactly where Triage operates rather than having it run across all boards.

Ignore Tickets Rules

Optionally list phrases that you do not wish to allow ServiceAI to triage. This is helpful for things like RMM alerts, backup reports and more - anything that you don't want to waste triage credits on.

You can add as many rules as you like by selecting the + Add Rule button.

Excluded Sources

Use the Excluded Sources multi-select to prevent Triage from processing tickets that originate from specific sources in your PSA. This is useful when certain ticket sources (such as RMM integrations, monitoring tools, or automated email addresses) create tickets that should not consume triage credits or be modified by AI.

How to configure:

  1. Navigate to Settings > Triage
  2. Scroll to the Excluded Sources field under Additional Configuration Options
  3. Select one or more sources from the dropdown — the list is populated from your PSA's available ticket sources
  4. Click Save Settings

Tickets arriving from any excluded source will be silently skipped by Triage. They will not appear in the Triage log, will not count against your monthly triage limit, and will not be modified in your PSA.

Tip: This works well alongside Ignore Ticket Rules (which filter by subject line content). Use Excluded Sources when entire ticket sources should be bypassed, and Ignore Ticket Rules when you need keyword-based filtering within sources you otherwise want triaged.

Spam Destination

Optionally specify a board or queue where spam tickets should be moved. Leave this blank to disable spam filtering. The AI will detect spam tickets and route them accordingly.

Spam Bypass for Known Senders

Tickets whose requester email address matches an existing user in your Tenant are automatically excluded from spam detection. This avoids false positives when end-users in your client environments submit tickets that happen to use phrasing common in spam (security alerts, password reset requests, etc.). Free email providers (gmail.com, outlook.com, etc.) are excluded from this match — only domains that you have users associated with are considered "known."

Automatic Field Updates

Select which ticket fields the AI is allowed to automatically update. Available options may vary based on your PSA, but typically include:

  • Subject
  • Category/Type
  • SubCategory/Subtype
  • Classification/Item/Configuration
  • Priority
  • Work Type
  • Estimated Time
  • Billing Code
  • Board/Queue (Routing)
  • Ticket Type (Autotask only)

Only enable the fields you want the AI to manage. This gives you granular control over what changes Triage can make. The Ticket Type option appears only for Autotask tenants.

The Automatic Field Updates allow-list is honored end-to-end: the Triage Details page, the AI's prompt, and the actual PSA writes all use the same enabled-field set. Fields you have not enabled are not shown as proposed changes in Triage Details, are not included in the AI's analysis prompt, and are not written back to your PSA — so you only ever see and act on field changes for fields you have explicitly allowed.

Priority Alert Keywords

Define keywords that trigger automatic priority escalation. When these keywords appear in a ticket's subject or description, the ticket will be flagged for priority handling.

Common examples include: bitcoin, breach, encrypted, hacked, malware, ransom, ransomware, virus

You can add custom keywords relevant to your business by entering them in the field (minimum 3 characters) and clicking + Add.

Viewing Triage Results

Once Triage is running (in either Testing Mode or live), you can view the results by selecting Triage from the left-hand sidebar. This area serves as a log of all triage decisions made by the AI.

Confidence-style columns in the Triage grid (Spam Confidence, Duplicate Confidence, Routing Confidence, Fast Feedback Confidence) all render as percentages.

Triage Details View

Selecting an individual triaged ticket shows you the complete decision-making process the AI followed:

  • Ticket Received: Shows when the ticket was received, along with the Ticket ID, Original Subject, Requester Name, Requester Email, and a summary of the issue.
  • Spam Detection: Displays whether the ticket was flagged as spam or marked as clean, including a confidence percentage and reasoning.
  • Routing Analysis: Shows any routing changes made (or that would be made in Testing Mode) based on your Routing Rules.
  • Duplicate Detection: Indicates whether the ticket appears to be unique or a duplicate of an existing ticket. Triage now also recognizes forwarded copies of an existing ticket as duplicates of the original — previously, an email submitted as a ticket and then forwarded to the support mailbox could result in two separate tickets without a duplicate relationship. The forwarded version is now linked back to the original ticket and surfaced in the Duplicate Detection section just like any other duplicate match.
  • Field Classification: The core of the triage decision — shows the before and after values for each field the AI updated. Only the fields you have enabled under Automatic Field Updates appear here; disabled fields are not shown as proposed changes and are not written back to your PSA.

Assignment Analysis: Displays any technician assignment decisions based on your Assignment Rules.

Processing Summary: At the bottom, you'll see processing time, current mode (Testing or Live), and which fields were configured for update. In Testing Mode, a note confirms that no changes were made to the ticket in your PSA system.

Ticket Persistence

Triaged tickets remain in the Triage log even if they're deleted from your PSA. This log specifically serves as a historical record to understand the AI's decisions at the time of processing.

Zendesk-Specific Triage Behavior

In addition to the unified rules above, Zendesk triage has a few integration-specific capabilities:

  • Ticket forms: ServiceAI auto-discovers the ticket forms configured in your Zendesk instance, along with each form's field layout. During triage the AI can select the appropriate form and update its form-specific custom fields in a single pass
  • Picklist auto-discovery: Standard and custom ticket field picklists, groups, and agents are refreshed automatically — no manual list maintenance is required
  • Comments: When comment creation is enabled, Triage posts its decision as a comment on the Zendesk ticket

Recommendations to Get Started with Triage

  1. Start with Testing Mode enabled to observe AI behavior without risk
  2. Begin with a limited set of monitored boards rather than all boards at once
  3. Add a few clear routing and categorization rules based on your most common ticket types
  4. Review the Triage log regularly to identify patterns and refine your rules
  5. Gradually expand your rule sets and monitored boards as you gain confidence
  6. Disable Testing Mode only when you're satisfied with the AI's decision-making

Triage is designed to reduce the manual effort of ticket categorization and routing while ensuring tickets get to the right place faster. By using natural language rules, you can customize the AI's behavior to match your team's specific workflows and preferences.

PSA Field Name Reference

ServiceAI uses the native field names from your PSA system. The table below shows how internal triage fields map to each supported PSA:

ServiceAI Field ConnectWise Autotask
Routing Board Queue
Category Type Issue Type
Subcategory SubType Sub-Issue Type
Classification Item Ticket Category
Subject Summary Title
Estimated Time Budget Hours Estimated Hours

These labels appear throughout the Triage Detail view, Triage Settings checkboxes, and Triage grid column headers.

Halo, Syncro, and Kaseya BMS triage uses each system's native field labels (Team, Type, Category, etc.). The exact field names appear in the Triage Detail view and in your Triage Settings checkboxes for each tenant.

Recovering a Spam-Flagged Ticket

If a ticket was incorrectly identified as spam and deleted, you can restore it directly from ServiceAI:

  1. Enable the Show Deleted toggle on the Tickets page to display spam-deleted tickets.
  2. Click on the ticket to open its detail view.
  3. Click the Include in AI button to restore the ticket.

Once restored, the ticket will be re-included in AI analysis. The spam flag is cleared so it will not be re-deleted on the next sync.

Note on PSA Internal Notes

ServiceAI no longer posts an internal note to your PSA system when a ticket is identified as spam. Spam detection is handled silently within ServiceAI to avoid cluttering your PSA ticket history.

Triage Rule Syntax: Common Mistakes

Unified Triage Rules in ServiceAI are written as plain natural-language sentences. They are not parsed as code, so programming-style expressions, comma-separated quoted lists, and if/then blocks will not be evaluated the way you might expect.

Avoid:

if contains File Share Event - External, File Upload Limit Exceeded, Stale Account Clean Up change queue to Network Alerts

Use instead:

Tickets with subjects containing File Share Event - External, File Upload Limit Exceeded, or Stale Account Clean Up should be routed to the Network Alerts queue.

When a triage rule is not firing as expected, double-check the following:

  • The rule is phrased as a natural-language sentence rather than a code-like expression.
  • The rule lives in the Unified Triage Rules list for your PSA.
  • The queue or board name in the rule exactly matches the queue or board name in your PSA.
  • Triage is not currently in Testing Mode.
  • The field you are trying to update is enabled under Automatic Field Updates.

Related articles

Comments

0 comments

Article is closed for comments.