Customer support sits on the largest unread dataset in the company. Tickets pile up in Zendesk, conversations sprawl in Intercom, interview notes drift in Productboard. Weekly synthesis usually means a CS lead reading for a day, then writing a memo nobody fully trusts. Dock runs synthesis as an agent shift: tickets and notes in, insights memo out, product owner attributed on the row. Every claim cites a source ticket.
Zendesk, Intercom, and Productboard 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 Insights Memo table
| Theme | Source tickets | Drafted by | Product owner | Status |
|---|---|---|---|---|
| Onboarding: SSO setup confusion | ZD#4412, ZD#4480, IC#9912 | mei-cs (agent) | Priya Shah | Reviewed, queued for roadmap |
| Billing: proration on plan downgrade | ZD#4501, ZD#4533, IC#9988, PB-note-318 | mei-cs (agent) | Daniel Okafor | Owner editing |
| Mobile: push notifications silent on iOS 18.4 | IC#10021, IC#10044, ZD#4612 | mei-cs (agent) | Priya Shah | Draft |
Each row links back. Priya can open ZD#4412 and read the original ticket. She sees mei-cs clustered three tickets under one theme, drafted the framing, and flagged her as owner because the area maps to her surface. The agent identity is not a service account. See agent identity for why that distinction matters when a memo influences the roadmap.
The workflow
Monday morning, mei-cs pulls the last seven days of resolved tickets from Zendesk and Intercom, plus interview notes tagged in Productboard. It clusters by theme, drops anything with fewer than three supporting tickets, and drafts a short memo per cluster: the pattern, the customer language, the suspected root cause, and three to five linked source tickets. It writes the row and assigns a product owner based on surface ownership.
The owner gets a Dock notification, reads the draft, clicks through two or three source tickets to sanity-check, then edits or marks it reviewed. Edits are tracked. The audit trail shows what the agent drafted, what the human changed, and when. See agent audit and compliance.
Why this matters
Synthesis by a person is a bottleneck. Synthesis by an agent without attribution is noise. Dock makes the agent's interpretation a first-class artifact: stored, cited, owned by a named reviewer. Productboard describes its Spark AI as a way to "uncover trends across large volumes of feedback" and "analyze thousands of inputs" with "20+ integrations" (Productboard, Spark). The capability is real, but the gap most teams hit is not analysis, it is trust. Pendo's State of Product Leadership found the majority of PMs see themselves as "tactically focused" and "more driven by competitors than by customer needs" (Pendo). A named reviewer on every memo changes that.
The agent does not own the decision. It drafts. Priya approves. The roadmap still belongs to a human, but the reading layer underneath it scales. See Cloud 2.0 for product.
Run the synthesis shift in Dock
FAQ
Who owns the memo if the agent drafted it? The product owner listed on the row. The agent is cited as the drafter, the same way a junior analyst would be cited, but the reviewer is accountable for what ships to the roadmap.
What if the agent miscategorizes a ticket? The product owner edits the row. The original draft is preserved in the audit trail so the next synthesis can learn from the correction. See the Dock for customer support overview for how this loop tightens over weeks.
Does the agent have its own login to Zendesk and Intercom?
Yes. mei-cs holds its own scoped API credentials, not a borrowed human session. This is the agent identity pattern in agent identity.
Can the memo be exported for a roadmap review? Yes. The row exports as markdown or PDF with all source-ticket links intact, so the document a PM brings to roadmap review is the same document the agent drafted, with the human edits tracked inline.
