Connector name:zendesk support

Table name: ticket_metrics

API documentation link: https://developer.zendesk.com/api-reference/ticketing/tickets/ticket_metrics/

Description / Justification:

We request adding native support for syncing the /api/v2/ticket_metrics.json endpoint directly as a ticket_metrics table within the Zendesk Support connector.

While Fivetran currently recommends calculating core KPIs downstream using raw tables and the dbt_zendesk package, both community feedback and our real-world production usage highlight severe functional and cost limitations with this approach:

• Inaccurate & Incomplete Metrics:

  • Messaging & Chat Replies (author_id = -1): In Zendesk Messaging/Chat, replies recorded under system placeholder IDs fail to match user tables, causing them to be filtered out in dbt. This leaves critical metrics like reply_time_in_minutes blank and results in incorrect reply counts.

  • Private Opening Messages: When tickets originate via API or web forms with a private initial comment (public = false), dbt skips the agent's actual first public reply, skewing response-time accuracy.

  • Pausable Update Time: Metrics such as pausable_update_time cannot be reliably recalculated from available raw data.

• Massive MAR Spike via Event Tables: To work around messaging discrepancies, Fivetran support recommends enabling event tables like TICKET_CHAT_EVENT. However, this generates upwards of 1 million rows per month, creating an unsustainable and cost-prohibitive spike in Fivetran Monthly Active Rows (MAR) just to fix a handful of reply metrics.

• Excessive Warehouse Compute Overhead: Recalculating these metrics continuously requires executing 60+ intermediate models in the dbt_zendesk package on an hourly basis, driving up database compute costs.

• Source-of-Truth Parity: Zendesk pre-computes native metrics (reply_time_in_minutes, reply_count, etc.) directly via API. Syncing this table directly ensures 100% parity with Zendesk Explore reporting while adding minimal MAR (1 row per ticket update) and eliminating downstream compute dependency.