Feature Configuration

Once your Zendesk instance is connected and active, you can enable, disable, and tune individual AI features from the Flexivity AI administration console. Each feature is configured independently per instance.

This page covers the ongoing tuning surface — the per-feature on/off toggles, prompt customization, classification topic-field setup, sync window selection, feedback settings, and Zendesk-side prerequisites. For the one-time connect flow (OAuth, marketplace install), see Connecting Zendesk.

What You Need for a Working System

For Flexivity AI features to run end-to-end on your Zendesk instance, four things must all be true:

1. The Flexivity AI marketplace app is installed on your Zendesk account — done in Marketplace Installation.
2. Flexivity is connected to Zendesk via OAuth — done in Connecting Zendesk.
3. The historical ticket database sync has completed — kicks off automatically the moment you grant OAuth. Monitor progress and tune the sync window in Historical Sync Period below.
4. A classification topic field is configured — required for AI Classification to set or suggest a topic value on incoming tickets. Configure in Classification Topic Source (Zendesk) below.

Items 1 and 2 are one-time setup steps. Items 3 and 4 are set up and monitored from this Feature Configuration page.

Accessing Feature Settings

1. Sign in to the Flexivity AI administration console .
2. Navigate to Instances and select your Zendesk instance.
3. Scroll down to the Features section.

Each feature has its own settings card with a toggle and configuration options.

Enabling and Disabling AI Features

Every AI feature has an Enabled toggle on its settings card. Toggling a feature off:

• Stops new ticket events from being processed for that feature. Webhooks continue to arrive but the feature is skipped.
• Does not delete historical data (past summaries, embeddings, classification records). Those remain in place and become visible again if the feature is re-enabled.
• Does not remove already-written data from Zendesk. If AI Classification previously set a topic value on a ticket, that value stays on the ticket until a human changes it.

Disabling a feature is reversible — re-enabling resumes processing for new ticket events going forward, with no need to re-sync or rebuild any state.

Tip: All AI features are enabled by default on a new connection — there is no need to turn them on incrementally. Leaving features enabled while the historical sync is still in progress, or before you have configured the classification topic field, is harmless: each feature simply waits for its prerequisites and starts producing output the moment they are in place. You do not need to time the toggles to match setup progress.

Prompt Customization

The AI prompt templates used for Summaries, Recommendations, Classification, and KB Article drafts can be customized per instance. This allows you to tailor the AI output to your organization’s tone, terminology, or specific instructions.

Accessing Prompt Settings

1. In the admin console, navigate to your Zendesk instance.
2. Open the Features section and find the feature you want to customize.
3. Click Configure prompt (or the equivalent action link on the feature card).

Editing a Prompt

Each feature has a default prompt template that works well for most use cases. To customize:

1. Click Edit on the prompt you want to modify.
2. Modify the prompt text. The editor shows available template variables (e.g., {{ticket_subject}}, {{ticket_body}}, {{thread_messages}}, {{topic_list}} for Classification).
3. Click Save Draft to save without activating.
4. Click Publish to make the new prompt active.

What’s Safe to Change vs What Breaks the Feature

• Safe: changing tone, adding domain-specific instructions (“our customers are healthcare professionals — avoid medical jargon”), adjusting summary length guidance, listing words to avoid.
• Risky: removing or renaming template variables — the feature will fail at runtime if a referenced variable is missing.
• Breaking: requesting structured output in a format Flexivity does not expect (e.g., asking the LLM to return XML when Flexivity parses JSON). The output will fail post-processing and the feature will show errors in the job log.

Note: Prompt changes take effect immediately for new events. In-flight events already queued for processing will use the prompt version that was active when they were enqueued.

Reverting a Prompt

Each prompt edit creates a new version. You can view the version history and revert to any previous version:

1. In the prompt editor, click Version History.
2. Select the version you want to restore.
3. Click Revert to This Version and then Publish.

The previously-active version becomes the live prompt. The reverted-from version remains in history so you can roll forward again if needed.

Classification Topic Source (Zendesk)

AI Classification on Zendesk writes to your existing Zendesk topic ticket field — you pick which custom field represents your taxonomy. Unlike the osTicket integration, Flexivity does not provision its own “Flexivity Topic” field.

Configuring the Topic Field ID

1. In Zendesk, find the custom field id for your topic field. Admin Center → Objects and rules → Tickets → Fields. Click the field you want to use; the URL contains the numeric field id (e.g., /admin/objects-rules/tickets/ticket-fields/360001234567 → field id 360001234567).
2. In the Flexivity admin console, open your Zendesk instance and navigate to Features → AI Classification → Configure.
3. Paste the field id into the Topic field id input and click Save.
4. Flexivity immediately fetches the field’s current dropdown options and seeds the picklist cache.

Picklist Cache Refresh Behavior

Flexivity caches the list of available topic values from Zendesk and refreshes it once per 24 hours automatically. New values added in Zendesk are not immediately eligible classification targets until the next refresh.

To force an immediate refresh after editing the field in Zendesk:

1. In the Flexivity admin console, open Features → AI Classification.
2. Click Refresh topic values now.

This synchronously re-fetches the dropdown options and updates the cache.

Behavior When the Field is Missing or Renamed

If you delete the configured field in Zendesk, or change its type away from dropdown, AI Classification will fail to write back and surface errors in the job log. Re-configure the field id to a new field — or disable AI Classification — to clear the errors.

Zendesk-side prerequisite: a topic dropdown field must exist

Zendesk-side prerequisite: A Zendesk custom dropdown ticket field representing your topic taxonomy must exist before AI Classification can be enabled. Create or open the field at: Zendesk Admin Center → Objects and rules → Tickets → Fields → Add field → Drop-down. For detailed instructions, see Zendesk’s documentation on custom ticket fields .

Auto-Assign Threshold

Classification has two modes — auto-assign and suggest-only — controlled by a confidence threshold (0.0 to 1.0). Default is 0.85.

• When the AI’s top-match confidence is above the threshold, Flexivity writes the topic field automatically.
• When the AI’s top-match confidence is below the threshold, Flexivity surfaces ranked suggestions in the widget for an agent to apply.

Lower the threshold to auto-assign more aggressively. Raise it to fall back to suggestions more often.

Historical Sync Period

The historical sync period sets how far back in your Zendesk ticket history Flexivity should pull on initial sync (and any re-sync).

To change the period:

1. In the Flexivity admin console, open the Zendesk instance and navigate to Features → Historical Sync.
2. Select the desired period from the dropdown.
3. Click Save, then Re-run Sync to backfill any newly-in-scope tickets.

Tip: Expanding the period costs LLM tokens (each new ticket gets embedded) and takes wall-clock time. Plan re-syncs during low-traffic windows. Subsequent webhook-driven processing is unaffected by re-sync.

Note: The window of solved tickets that KB Analysis looks at on each run is not configured here — an administrator selects it at the time they kick off a gap analysis. See User Guides — KB Analysis & Authoring (Zendesk) for the analysis workflow.

KB Analysis & Authoring

KB Analysis clusters your solved tickets, identifies gaps relative to your published Help Center articles, and (on demand) generates draft articles you can push back into Zendesk Help Center for admin review and publish. Three settings on this page tune how the analysis runs and how its drafts are written.

Article Format

When KB Analysis generates draft articles for the Zendesk Help Center, it can produce them in two styles:

Pick the format that matches how your Help Center is structured today — the AI uses the same template for every draft generated on the instance.

Coverage Threshold

The coverage threshold is the minimum similarity score required for an existing Help Center article to be considered a match for a ticket cluster.

• Default: 0.80.
• Lower the threshold to mark more clusters as covered — fewer gap candidates surface.
• Raise it to require closer matches before a cluster is considered covered — more gap candidates surface.

If your existing Help Center articles already cover most of the patterns the analysis finds, you may need to raise the threshold to surface meaningful gaps.

Minimum Cluster Size

The minimum cluster size is the smallest number of similar tickets required before a topic forms a cluster the analysis will look at.

• Default: 5.
• Higher values focus the analysis on frequently-recurring topics — the highest-impact gaps.
• Lower values catch niche issues that may still warrant an article, at the cost of statistical significance per cluster.

Changing These Settings

1. In the Flexivity admin console, open the Zendesk instance and navigate to Features → KB Analysis.
2. Adjust the Article format, Coverage threshold, or Minimum cluster size as needed.
3. Click Save.

The next gap-analysis run picks up the new settings. Drafts already pushed to Zendesk Help Center are not retroactively rewritten — re-generate a draft from the same cluster if you want it under the new format or to reflect a refreshed threshold.

See User Guides — KB Analysis & Authoring (Zendesk) for how these settings affect the analysis workflow and draft content.

Zendesk-side Prerequisites Summary

Several Flexivity features depend on Zendesk-side configuration that the Flexivity admin console cannot set up for you. Here is the consolidated list:

1. Zendesk Suite Team plan or higher

Zendesk-side prerequisite: Your account must be on Zendesk Suite Team plan or higher. Check at: Zendesk Admin Center → Account → Subscription. For plan details, see Zendesk’s pricing page .

2. Zendesk Support module enabled

Zendesk-side prerequisite: Zendesk Support must be enabled (required for tickets, comments, webhooks). This is included with Zendesk Suite plans by default and does not need separate enablement on those plans. For more detail, see Zendesk Support overview .

3. Zendesk Help Center module enabled

Zendesk-side prerequisite: Zendesk Help Center must be enabled (required for KB Analysis & Authoring and for Help Center context in Recommendations). Enable at: Zendesk Admin Center → Help Center → Settings. For setup instructions, see Zendesk’s Help Center setup guide .

4. OAuth scopes granted (read + write)

Zendesk-side prerequisite: During connect, the connecting admin must grant both read and write scopes on the OAuth consent screen. Granted at: the OAuth consent screen during the Flexivity connect flow. If revoked, manage at: Zendesk Admin Center → Apps and integrations → Connected apps. For more on Zendesk OAuth, see Zendesk’s OAuth documentation .

5. Topic ticket field exists (required for AI Classification)

Zendesk-side prerequisite: A Zendesk custom dropdown ticket field representing your topic taxonomy must exist. Create or open at: Zendesk Admin Center → Objects and rules → Tickets → Fields → Add field → Drop-down. For detailed instructions, see Zendesk’s documentation on custom ticket fields .

6. Connected user has Help Center publish permission (required for KB Authoring)

Zendesk-side prerequisite: The Zendesk user that authenticates during OAuth connect must have permission to create draft articles in your Help Center section. Grant at: Zendesk Admin Center → Help Center → Settings → Permissions. For permission setup, see Zendesk’s Help Center permissions guide .

Feature Health Monitoring

The Flexivity admin console provides operational metrics for each instance on the Dashboard tab:

• Jobs Today — Number of AI processing jobs executed in the current day.
• Failure Count — Number of failed jobs. A non-zero count warrants investigation.
• Connection Status — Current OAuth state (Active, Reauthorization required, Disconnected).
• Last Event Processed — Timestamp of the most recent successfully processed ticket event.
• Webhook Health — Indicates whether the Zendesk-side webhook subscription is still active and delivering 2xx responses.

If the failure count is elevated, check the instance’s job history for error details. Common causes include transient cloud errors (which are retried automatically) and OAuth refresh failures (see Troubleshooting).

Related Pages

For agent-facing documentation on each feature, see:

• User Guides — AI Summaries (Zendesk)
• User Guides — Recommendations (Zendesk)
• User Guides — AI Classification (Zendesk)
• User Guides — Feedback (Zendesk)
• User Guides — KB Analysis & Authoring (Zendesk)

For data handling and retention, see Data Disclosures — Zendesk.