AI Classification

AI classification automatically assigns or suggests the right topic for new tickets as they come in. Instead of relying on requesters to pick the correct category — or agents to manually retag — the AI analyzes the ticket content and matches it to the most appropriate topic value.

What Does Classification Do?

When a new ticket is created, Flexivity AI reviews the ticket subject and first comment to determine which topic value it belongs to. Depending on your configuration, it either:

Auto-assigns the topic field when the AI is highly confident in its match.
Suggests ranked topic values for an agent to review and apply with one click.

This reduces misrouted tickets, speeds up triage, and ensures tickets reach the right team faster.

How Classification Maps to Your Zendesk Topic Field

The Zendesk integration reads from and writes to your existing topic ticket field. You pick which Zendesk custom field represents your taxonomy (a dropdown field, typically called “Topic”, “Category”, or something specific to your business), and your administrator configures Flexivity to use that field by its custom-field ID.

This means:

The topic values Flexivity classifies into come from your Zendesk field’s existing dropdown options. If you have a topic value called “Billing — Refund Requests”, that is the exact value Flexivity writes.
You manage the taxonomy in Zendesk. Add, rename, or remove topic values in Zendesk’s ticket-field admin, and Flexivity will pick up the changes (with a brief cache refresh — see Feature Configuration).
Classifications appear in the field your team already uses. Flexivity does not add a separate field to your tickets.
Your existing Zendesk automations keep working. Any triggers, automations, SLA policies, or routing rules your admin has built on the topic field continue to fire as designed — Flexivity is just streamlining the step that decides which topic value gets set.

Tip: If your existing topic field is well-structured (clear labels, mutually-exclusive values, no duplicates), AI classification will be more accurate from day one. If the field is messy, it is worth cleaning up before enabling classification.

Where Does It Appear?

Classification appears in two places:

The Zendesk ticket itself — when classification auto-assigns, the topic field value is set directly on the ticket. You see it in the Zendesk ticket sidebar’s standard fields area, alongside any other custom fields you have configured. Nothing about the standard Zendesk UI changes.
The Flexivity AI widget — when classification runs in suggest mode (or when the AI’s confidence is below the auto-assign threshold), the widget shows a ranked list of suggested topic values. Each suggestion has an Apply button that sets the topic field with one click.

If the ticket already has the right topic value (because classification auto-assigned, or you set it manually), the Classification section may be hidden to reduce widget noise.

How Does It Work?

Classification uses a two-step process:

Similarity search — The AI compares the new ticket’s content against your team’s historical Zendesk tickets (the ones the historical sync has embedded) to find the most similar ones and their topic values. This is based on the meaning of the content, not just keyword matching.
LLM tiebreak (when needed) — If the similarity search produces multiple strong candidates, or if no historical ticket has a clear winner, an LLM step compares the ticket’s intent against the topic-value labels (and your administrator’s internal descriptions) to determine the best fit.

The result is a ranked list of topic values with confidence scores, which Flexivity then either applies or surfaces depending on your administrator’s auto-assign threshold.

Two Modes: Auto-Assign and Suggest

Your administrator configures how classification behaves based on a confidence threshold:

Auto-assign mode — When the AI’s confidence in the top match exceeds the threshold, it automatically writes the topic field on the ticket using the Zendesk API. No agent action is required.
Suggest mode — When confidence falls below the threshold (or when suggest-only mode is enabled), the AI presents ranked suggestions in the Flexivity widget. You review the options and click Apply on the one that fits.

In both modes, the agent always has the ability to override the classification and set a different topic value manually using Zendesk’s standard field editor.

Note: The confidence threshold is configured by your administrator in the Flexivity AI admin console. If you feel auto-assignments are too aggressive or too conservative, let your administrator know.

Accuracy Tracking

Flexivity AI tracks classification accuracy over time. When a ticket is solved, the system compares the final topic value on the ticket with the value Flexivity originally classified. This comparison feeds into accuracy metrics that your administrator can review in the analytics dashboard.

This means:

If the AI got it right, that success is recorded.
If an agent changed the topic value before solving, the system records the correction as a learning signal.

Over time, these metrics help administrators understand how well classification is performing and whether the confidence threshold needs adjustment.

Never Undo Human Work

Flexivity AI checks the ticket’s audit history before applying or re-applying a classification. If a human agent has already set or changed the topic field after Flexivity’s session started, Flexivity will not overwrite that change — your decision sticks.

This means once you set the topic value manually on a ticket, automated reclassification on subsequent updates is skipped for that ticket. Your edit is treated as the source of truth.

Tips for Getting the Most Out of Classification

Classification improves over time. As your team solves more tickets, the AI has more data to learn from. Early on (especially right after the historical sync completes), you may see more suggestions than auto-assignments. This is expected.
Correct misclassifications before solving tickets. If the AI assigned the wrong topic value, change it to the correct one before solving. This correction is captured in accuracy tracking and helps the system improve.
Classification only runs on new tickets. Existing tickets that were created before classification was enabled are not retroactively reclassified. The feature applies to incoming tickets going forward.
Keep your topic field tidy. Removing dead topic values from the Zendesk dropdown helps the AI focus on the ones that are still in use.

Troubleshooting

I do not see any classification on a ticket. Classification only runs on new tickets. If the ticket was created before the feature was enabled, no classification will appear. Additionally, if the AI could not find any matches with sufficient confidence and no auto-assignment was made, the classification suggestions may not appear in the widget.

The auto-assigned topic value is wrong. Change the topic field manually to the correct value using Zendesk’s standard field editor. This correction is tracked and helps improve future classifications. If incorrect auto-assignments happen frequently, let your administrator know — the confidence threshold may need to be raised.

I only see suggestions, never auto-assignments. Your administrator may have set a high confidence threshold or configured the system to always show suggestions rather than auto-assign. This is a deliberate choice to keep agents in control of routing decisions.

A topic value I just added in Zendesk is not appearing as a suggestion. Flexivity caches the list of available topic values from Zendesk and refreshes it periodically. New values may take a brief period (typically up to 24 hours, or sooner if your admin forces a refresh) to become eligible suggestions. See Feature Configuration — Zendesk for how to trigger a refresh.

What Your Administrator Can Configure

The classification prompt, the topic field id, the auto-assign threshold, the picklist refresh, and the on/off toggle are all configurable — see Feature Configuration — Zendesk. For what data Flexivity reads from your tickets to classify them, see Data Disclosures — Zendesk.