Case Study
Fijord: From Meeting Chaos to Shipped Tickets
Designing an AI-powered product management tool that transforms discovery calls into evidence-backed work items.
The Problem
“When I wasn’t in that meeting, I didn’t have full context on why we’re even doing this.”
— Product Manager, Amazon
Product managers spend hours after every user interview doing tedious work: re-watching recordings, copying quotes into documents, synthesizing findings, and writing tickets manually. The real cost is that the evidence trail disappears. By the time a ticket reaches engineering, the original user quote is gone.
The Solution
Fijord processes meeting transcripts and extracts problems backed by evidence, then generates tickets that can be exported directly to Linear or Jira. Every ticket maintains a link back to its source quotes.
Process a Transcript
Users paste a transcript or connect Fireflies.ai to import directly. The input is intentionally simple: one text area, one button.
Evidence, Problems, and Tickets
AI processes the transcript in 14 seconds and extracts 12 quotes, 3 problems, and 19 suggested tickets. The three-column layout lets users see the full chain from evidence to action.
Trace Back to Source
Click any quote to open the transcript drawer with the highlighted context. The transcript is accessible when needed for verification, but doesn’t dominate the interface.
Select and Stage
Users explicitly select which tickets to keep. This intentional friction ensures PMs review AI suggestions rather than blindly exporting everything.
Organize Before Export
A simple kanban staging area for prioritization and editing before exporting to Linear or Jira.
Full Evidence Trail
Each exported ticket preserves its lineage: problem statement, supporting quotes, acceptance criteria, and a shareable backlink to Fijord. When engineering asks “why are we building this?”, the PM can point to the source.
Design Process
Early versions showed tickets first. But user testing revealed that PMs didn't trust AI-generated tickets without seeing the source. By leading with evidence — the actual user quotes — users can verify the AI's reasoning before accepting its suggestions.
The three-column layout mirrors the thinking process: What did users say? → What problems does that reveal? → What should we build?
The evidence-first layout: quotes → problems → tickets
Outcomes
Designed & in development. Full transcript-to-tickets flow, Linear/Jira integrations, backlinks, and Signals system.
“That full lineage from messy call data to the ticket, that’s what speaks to me.”
— Amazon PM
Evidence before action
Users don't trust AI outputs without seeing the source. Show the quotes first.
Backlinks are the moat
Anyone can extract text from a transcript. Preserving the evidence trail is what makes this valuable long-term.
Design engineer velocity
Being able to implement my own designs cut iteration cycles from days to hours.