Small support teams live in Help Scout. The mailbox holds the conversation, the customer profile, and the saved replies. What it does not hold is a structured record of what the agent thought, what it almost said, and which editor approved the final reply. Dock adds that layer. The agent reads the mailbox, drafts a response brief in a Dock table, and waits for a named editor to approve before the message ships. The conversation stays in Help Scout. The interpretation stays in Dock.
Help Scout and Mailbox 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 response brief table
One Dock table, cs.response-briefs, holds every drafted reply before it ships. Each row is the agent's interpretation of a single Help Scout conversation.
| conversation_id | customer | intent | proposed_reply | confidence | drafted_by | editor | status |
|---|---|---|---|---|---|---|---|
| HS-48201 | maya@northbend.co | refund_window_exception | "We can extend your return window to June 14. Sending the prepaid label now." | 0.82 | echo@dock | nadia@team | approved 14:22 |
| HS-48207 | raj@vetterly.io | bug_report_checkout | "Logging this with engineering. I will update you within one business day." | 0.91 | echo@dock | nadia@team | approved 14:31 |
| HS-48214 | ops@brightline.com | bulk_pricing_question | "Pricing for 50+ seats needs sales review. Routing to Govind." | 0.64 | echo@dock | nadia@team | escalated 14:38 |
The proposed reply is text the editor can edit in place. The status column gates the Help Scout send action. Nothing leaves the mailbox until a human row owner flips status to approved.
One workflow, end to end
A new conversation lands in the support mailbox. The agent (Echo, signed in with its own identity) pulls the thread, the customer's prior tickets, and the relevant help-doc snippets. It writes a row to cs.response-briefs with intent, proposed reply, and a confidence score. Nadia, the editor on shift, opens the table view filtered to her queue. She tweaks two sentences in the proposed reply, sets status to approved. The agent re-reads the row, calls the Help Scout API to post the reply as a note attributed to Echo, and tags the conversation with the Dock row ID. If status flips to escalated, the agent assigns the Help Scout conversation to the named human instead. Every step writes to the row's audit log, which feeds the broader agent audit trail.
Why this matters
Help Scout reports that 60 percent of customers define an immediate response as ten minutes or less. Small teams cannot staff that bar without help. An agent drafting in parallel with a single editor reviewing can. The structure also fixes the harder problem: knowing later who said what and why. The Dock row is the receipt. The editor's name is on it. The agent's identity is on it. When a reply is wrong, you can find the moment it went wrong without scrolling a mailbox.
This is the same pattern we describe in how to run customer support with AI and the response drafting primitive. The editor lane is what makes a two-person team behave like a five-person team. The audit row is what keeps the agent inside its lifecycle boundaries.
Spin up the response-brief table in your workspace and connect your Help Scout mailbox. The first reply ships through Dock in under an hour. See the full pattern in our Dock for customer support guide.
FAQ
Does Dock replace Help Scout? No. Help Scout remains the mailbox, the customer profile, and the channel. Dock holds the agent's draft, the editor's approval, and the audit row.
What if the editor is offline? The row stays in pending status. The agent does not send. You can configure a fallback rule that escalates pending rows older than thirty minutes to a backup editor or to a human queue in Help Scout.
Does the customer see the Dock row? No. The customer sees only the final reply posted to the Help Scout conversation. The Dock row is internal and includes a pointer back to the conversation ID for traceability.
How does the agent handle a conversation it has seen before? It re-fetches the Help Scout thread on each pass rather than caching. Help Scout sees 78 percent of consumers expect responses within an hour, and a stale cache produces stale replies. Fresh reads keep the brief current.
