Ten Granola recordings into one PRD the team actually believes
Discovery calls stop dying in transcript folders. Recordings become themes, themes become a cited PRD, and the stakeholder brief writes itself from the same evidence.
- Steps
- 5
- Tools
- 7
- Metrics
- 3
- Read
- 4 min
A twelve-person Series A team with one PM and a designer who runs half the calls. Roughly fifteen customer conversations a month, recorded across three different tools depending on who booked them.
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
Three recording tools, three transcript formats, one folder nobody opened twice.
Synthesis was a two-day job that got skipped whenever a quarter got busy — so the last real synthesis was in March.
The PRD cited “customer feedback” with no way to check which customer, or whether the quote survived paraphrasing.
Engineering pushed back on the spec and nobody could produce the evidence fast enough to settle it.
Every recording lands in the same graph. Themes emerge across tools and months, not per-call. The PRD’s claims are clickable, so the argument in the review meeting is about the trade-off, not about whether the evidence is real.
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.
Granola ships its own hosted MCP server, so an agent can read meeting notes from one MCP endpoint and write the resulting theme into Specky’s — no export step, no connector to build.
Sales and success calls sync automatically with transcripts attached.
Recorded product-discovery calls sync as transcript nodes.
Runs autonomous JTBD interviews via a shareable link, so a slow interview month still produces evidence.
Clusters transcripts into JTBD themes with verbatim quotes, then drafts the PRD with each requirement cited.
The PRD and the stakeholder brief publish out to the team wiki where everyone already reads.
Theme digests posted to #product so research is ambient instead of an event.
How it actually runs
Every step is labelled by who does it. The steps marked 'you review' are the ones worth keeping human.
All three recorders feed one pool
AutomaticGong and Zoom sync natively; Granola exposes its own MCP server, so an agent can read from it and write into Specky’s without an export step. What matters is that a theme can span a Gong sales call from January and a Granola discovery note from July — which is exactly the pattern per-tool analysis can never see.
Alex fills the gaps you didn’t have time for
AgentLaunch a JTBD campaign as a shareable link and Alex interviews respondents autonomously. Themes, sentiment, and NPS responses land in the same graph as the human-run calls, so a light interview month doesn’t become an evidence hole.
Themes, with the quote attached
AgentSpecky clusters the pool into Jobs-to-be-Done themes, each backed by verbatim quotes rather than a paraphrase. You can see how many distinct accounts are behind a theme before you commit a quarter to it.
The PRD cites its way through the argument
You reviewGenerate the PRD from a theme and every requirement carries an inline citation to the sentence that justified it. When engineering asks “who asked for this?”, the answer is one click, in the doc, in the meeting.
The stakeholder brief comes free
You reviewThe same graph generates the exec-readable version — what we learned, what we’re doing, what evidence backs it — published to Notion and posted to Slack. You wrote it once, in the form of approving a PRD.
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.
Time from last interview to approved PRD
Same week — not “when I get a clear two days”
Transcript node created_at → PRD approved_at. If it’s still measured in weeks, synthesis is still the bottleneck.
Citation density
Every requirement traceable to at least one verbatim quote
Cited requirements ÷ total requirements per PRD. Specky renders this inline, so it’s visible during review rather than audited afterwards.
Distinct accounts behind a shipped theme
More than one, always
Count unique source accounts in a theme’s evidence set. This is the cheapest guard against shipping your loudest customer’s roadmap.
Start here, not everywhere
Connect the recorder you use most, then stop
Gong or Zoom natively; anything else gets forwarded in. The value shows up when a theme spans two tools and four months, and that only needs one connected source plus a habit of forwarding the rest.
Launch one Alex campaign this week
A shareable JTBD link costs you ten minutes and produces interview evidence while you sleep. It’s the fastest way to find out whether your theme pool is thin.
Publish the brief where people already read
Push the stakeholder version to Notion and post the digest in Slack. Research that lives in the research tool is research nobody uses.
https://www.specky.space/api/mcpClaude Code · Codex · Cursor · Claude Desktop · ChatGPT · n8n
Frequently asked questions
Does Specky integrate with Granola?+
Not with a first-party connector, and it needs one less than you’d think — Granola runs its own hosted MCP server and so does Specky, so an agent connected to both reads the meeting notes from one and writes the theme into the other. Failing that, forwarding or an automation hop into the capture API works and indexes identically. Gong and Zoom sync natively.
Can Specky replace our research tool?+
For continuous discovery, largely yes — Alex runs JTBD interviews autonomously and themes flow straight into specs. For moderated usability testing with a facilitator watching a screen, keep the tool you have and forward the output in.
How are citations verified?+
A citation points at a graph node — the actual transcript, message, or ticket — not at a paraphrase. Clicking it opens the source text. If a claim has no node behind it, it renders as uncited, which is deliberately hard to miss during review.
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