From a Tally form to a merged PR, overnight
A four-person team wires their feedback forms, inbox, and support chat into Specky, wakes up to an evidence-cited PRD, and hands it to Claude Code — which opens the PR.
- Steps
- 5
- Tools
- 7
- Metrics
- 3
- Read
- 4 min
A four-person B2B SaaS team. Two engineers, one founder doing product, one doing support. No PM, no research function, ~400 pieces of user feedback a month across three inboxes.
Composite blueprint. This is the loop and the wiring — assembled from workflows our early design partners run — not a single named customer’s audited results. Every number on this page is a metric to instrument, not a result we’re claiming.
What the week looked like
Feedback landed in four places — a Tally form, the Intercom inbox, support@, and DMs — and nobody owned reading all four.
The founder skimmed for the loudest complaint on Sunday night and turned it into a Linear ticket with two lines of context.
Engineers built the ticket, not the problem. Half the fixes missed what the user actually meant.
Nobody ever went back to tell the person who reported it that it shipped.
Feedback arrives, gets clustered against everything else ever said, becomes a cited PRD by morning, and a coding agent opens the PR against it. The human job shrinks to one thing: approving the right problem.
The tools this loop runs on
Native connectors where they exist, and an honest label where they don't — anything else reaches Specky over MCP, the REST API, or an automation hop.
The in-app 'what’s missing?' form. An n8n webhook posts each response to Specky’s capture endpoint.
Support conversations sync into the Product Graph as signals, tagged by account.
Forwarding rule → n8n → Specky, so the shared inbox stops being a black hole.
Clusters every signal into themes, scores the opportunity, drafts the PRD with inline citations, generates tickets.
Reads the approved PRD and its cited evidence over MCP, writes the change, opens the PR.
PR lands with the Specky doc linked in the description. Merged commits sync back as delivery signals.
Tickets pushed from the PRD with acceptance criteria intact.
How it actually runs
Every step is labelled by who does it. The steps marked 'you review' are the ones worth keeping human.
Every channel becomes one stream
AutomaticTally responses, Intercom threads, and forwarded support mail all land as typed nodes in the Product Graph. No triage rota, no spreadsheet, no 'who’s got the inbox this week'.
Overnight, the noise becomes a theme
AgentSpecky’s overnight agent clusters the week’s signals, notices that nine separate complaints are the same problem stated five ways, and scores the opportunity against everything already in the graph.
You get a PRD, not a summary
You reviewThe PM Inbox has a drafted PRD with a confidence score. Every claim is a citation — click it and you’re reading the exact sentence a real user typed. You approve, edit, or kill it.
Claude Code picks it up over MCP
AgentPoint Claude Code at Specky’s MCP server. It reads the approved PRD and its evidence nodes, writes the implementation, and opens a PR that links back to the spec — so the reviewer can see the user quote that justified the change.
Merge closes the loop
You reviewThe merged commit syncs back into the graph and attaches to the opportunity that started it. Specky can then draft the “you asked, we shipped” note to every account whose feedback fed the theme.
The numbers to watch — not the ones we'd like to quote
We're not going to invent a percentage for you. These are the three metrics that tell you whether the loop is real, and how to compute each one from data the graph already holds.
Feedback-to-fix cycle time
Days, not quarters — measure the median, not your best week
Timestamp on the first signal in a theme → merge timestamp on the PR that closes it. Both live in the graph, so the number is queryable rather than guessed.
Share of shipped work traceable to a user signal
The number you can defend in a board meeting
Count merged PRs linked to an opportunity with at least one cited evidence node, over all merged PRs.
Close-the-loop rate
Every reporter hears back
Accounts notified after a shipped fix ÷ accounts whose feedback fed that theme. This is the one that moves NPS, and it’s the one everybody skips.
Start here, not everywhere
Point one form at Specky first
Don’t wire everything. Take the single highest-volume feedback source you have and route it in — native connector if there is one, otherwise an n8n webhook to the capture endpoint. One source produces a real theme within a week.
Connect your coding agent to the MCP server
Add https://www.specky.space/api/mcp to Claude Code, Codex, or Cursor with a workspace API key. The agent can then search the graph, read the PRD, and cite evidence — instead of you pasting context into a chat window.
Keep the approval gate
Specky drafts in the PM Inbox and MCP tools create in draft state by design. Let the agent write the PR; keep a human on the merge. The speed comes from removing the writing, not the deciding.
https://www.specky.space/api/mcpClaude Code · Codex · Cursor · Claude Desktop · ChatGPT · n8n
Frequently asked questions
Does Specky write the code?+
No. Specky decides what’s worth building and writes the evidence-cited spec. Your coding agent — Claude Code, Codex, Cursor — reads that spec over Specky’s MCP server and writes the code. Specky is the layer that stops the agent building the wrong thing quickly.
Is there a native Tally integration?+
Not today. Tally reaches Specky through an n8n, Zapier, or Make webhook into the capture API, which takes about ten minutes to set up. Typeform, Intercom, Zendesk, Slack, Jira, Linear, GitHub, Gong, HubSpot, and PostHog have native connectors.
What stops the agent from shipping a bad change?+
The approval gate. Specky creates PRDs and tickets in draft state in the PM Inbox, and the coding agent opens a PR rather than pushing to main. A human approves the problem and reviews the diff — two gates that cost minutes and catch the expensive mistakes.
Run this loop on your own product.
One source, one agent, one approval gate. You’ll know inside a week whether it holds.
14-day free trial · No credit card required · Cancel anytime