Problem
Customer conversations were happening, but the learning risked becoming anecdotal, hard to compare, and too easy to lose when product, GTM, and fundraising decisions competed for attention.
Product management proof
Building A Customer-Intelligence Learning Loop
Across 100+ customer conversations and user interviews, I designed the learning loop, conducted interviews, built an n8n transcript-synthesis workflow, and handed source-grounded insights to Pier's technical cofounder.
What I owned
I routed transcripts by conversation type, designed structured extraction prompts, added schema validation and human review, and connected the resulting insight repository to product, positioning, and engineering decisions.
Result
Pier gained a reusable customer-intelligence system instead of relying on scattered transcripts, one-off notes, or one person's memory of each call.
Customer details and source records have been generalized to protect private relationships.
What I noticed
Pier was building an AI-assisted product for qualification and evaluation workflows. Employers and other workflow participants could provide rich signal about what evidence they trusted, where decisions stalled, and what the product needed to make clearer. Raw transcripts, however, could not become the team’s memory or roadmap.
The product-management problem was not producing more notes. It was designing a learning loop that could preserve the source, distinguish conversation types, extract decision-relevant signal, and move that signal into product, positioning, and engineering work.
At a glance
- Product area: customer discovery, product positioning, and MVP input for an AI-assisted qualification and evaluation product
- Users: employers and other workflow participants across several organization types
- Collaborator: Pier’s technical cofounder, who owned core-product engineering
- Constraint: interview learning was distributed across transcripts and working notes while product and positioning decisions were moving
- Decision: build a source-grounded learning loop with structured extraction, schema validation, human review, and a versioned handoff
- Artifacts: interview guides, extraction prompts, n8n workflow, schema validator, structured insight repository, and technical-cofounder handoff
- Observable result: customer learning became inspectable and reusable in product, positioning, and MVP feature decisions
What I owned
I designed the learning loop and conducted the interviews. I then built the technical path that turned transcripts into reviewable customer intelligence: conversation-type routing, a first LLM pass for quote-grounded extraction, a second pass that compiled the findings into a stable YAML structure, schema validation, valid-or-invalid routing, and human review before the result could change a source-of-truth document.
I also organized the accepted outputs in a versioned repository and handed the insights to Pier’s technical cofounder for MVP feature decisions.
The product decision
The decision was not to produce one confident summary from every call. A user interview, buyer conversation, and partner discussion were different evidence types, so they needed different extraction prompts and review standards.
The workflow prioritized source traceability over fluent prose. Each retained finding needed direct transcript support, and outputs that failed the schema or review standard stayed out of the learning repository. That made the handoff inspectable rather than turning customer language into an untraceable list of feature requests.
The technical system
The operating version used webhook-triggered orchestration to start transcript processing. The saved n8n export shows the core analysis path: transcript intake, quote-grounded extraction, YAML compilation, schema validation, and valid-or-invalid output routing. HubSpot fields and the versioned insight repository made the accepted learning reusable in the team’s operating work.
Result
Across 100+ customer conversations and user interviews, Pier gained a repeatable way to turn raw transcripts into source-grounded customer intelligence. The accepted insights were organized for reuse and handed to the technical cofounder as input to product, positioning, and MVP feature decisions.
What I learned
Customer discovery becomes product work when another person can inspect the signal, challenge the interpretation, and use it in a real decision. Technical automation made that loop faster and more consistent; source grounding and human review kept the decision accountable.