Zendesk holds the conversation. It does not hold the reasoning an agent applied when it routed a ticket to billing instead of churn risk. Dock does. The Zendesk ticket stays canonical for the customer-facing thread. A Dock row stores the agent's classification, the proposed macro, the analyst who signed off, and the timestamp. When a CS leader asks why a refund ticket sat in tier-one for nine hours, the answer is one row away. This is the customer support pillar applied to a Zendesk and Intercom stack.
Zendesk and Intercom stay the system of record for the raw data. Dock is the system of record for what the AGENT INTERPRETS. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The triage brief table
One Dock table per inbox. The agent writes a row per ticket it touches.
| Zendesk ticket | Agent classification | Proposed macro | Confidence | Analyst sign-off | Timestamp |
|---|---|---|---|---|---|
| #48211 (refund, 9h old) | Billing dispute, low churn risk | macro_refund_under_50 | 0.91 | sarah.k approved | 2026-05-30 09:14 |
| #48207 (login, repeat) | Auth bug, escalate to T2 | macro_t2_auth_handoff | 0.78 | rohan.p edited reason | 2026-05-30 09:02 |
| #48199 (cancel) | Churn save, retention path | macro_save_offer_3mo | 0.64 | sarah.k rejected, manual reply | 2026-05-30 08:47 |
The ticket column links back to Zendesk. The analyst column links to a named agent identity or a human. Confidence under 0.70 forces a sign-off field before the row can close.
Worked workflow
A refund ticket lands in Zendesk at 09:11. The Dock agent fetches it via API, reads the conversation, checks the customer's plan and lifetime value through a fresh Intercom read, and writes row #48211. It proposes macro_refund_under_50 at 0.91 confidence. Sarah, the CS analyst on shift, sees the row in her Dock queue. She opens the linked Zendesk ticket, agrees with the classification, and clicks approve. The macro fires in Zendesk under Sarah's name. The row closes with both the agent's reasoning and Sarah's sign-off attached. If the customer replies angry, the agent appends a new row pointing at the same Zendesk ticket. The original triage stays intact for review.
Why it matters
Zendesk's 2026 CX Trends report found that 95% of consumers expect explanations for AI-generated decisions and only 37% of companies currently provide them. The Dock row is that explanation, captured at the moment of decision, attributable to a specific agent identity. Zendesk's own QA data shows customers like Kahoot lifted ticket review volume 150% with structured QA; the bottleneck is reviewer time, not reviewer intent. Attributed agent drafts shift the analyst from author to approver. For the broader pattern, see how to run customer support with AI and the audit and compliance model that backs it.
Try it
Point a Dock agent at one Zendesk view. Give it a triage table. Require analyst sign-off for one week. Read the rows.
FAQ
Q: Does the agent reply directly in Zendesk? No. The agent drafts the macro in Dock. A named human or a separately credentialed reply agent fires it in Zendesk. The reply is always attributable.
Q: What happens when the agent is wrong? The analyst rejects the row, writes a reason, and replies manually. The rejected row stays. Weekly review surfaces patterns. Retraining is grounded in real rejections, not vibes. See agent identity lifecycle.
Q: Do we need Intercom and Zendesk both? No. One inbox works. Two inboxes mean one Dock table per inbox, same schema. The agent identity carries across.
Q: How does this differ from marketing triage? Same architectural shape, different platform. The Dock-row-as-interpretation pattern reappears in Dock for marketing and across every GTM rail.
