Sift
Product managers sit on hundreds of customer voices a month and still lose the roadmap argument to the loudest person in the room. Sift turns the voices into evidence they can defend.
- The problem
- Feedback lives in three to seven tools. Synthesis is done by hand, so a roadmap call cannot be traced back to the customers who asked for it. Every competitor solves collecting feedback. None solves making sense of it.
- The goal
- Make synthesis the home screen and every theme traceable to real words, so a product manager can defend a call in the room. The pricing bet: $30 to 40 a seat, with 8 to 12 percent of free teams converting.
- What I did
- Seven phases of research on live competitor pages, a journey built on 19 verbatim customer quotes, three personas, the jobs, an architecture of 39 specified pages and a 17-screen grey prototype with every state.
- What it gave
- A product specified end to end before any colour: every screen traces to a job, every job to evidence, every fact to a link. In the prototype the customer's own words are never more than two taps from a ranked theme.
Evidence loses to the loudest person in the room.
A product manager at a B2B software company of 50 to 500 people has feedback in Intercom, in Zendesk, in call recordings, in review sites and in a folder of interview notes. There is no shortage of tools that gather it. The moment that hurts comes later, in the room: a stakeholder asks why this is on the roadmap and not that, and the honest answer is a feeling.
The current journey bottoms out exactly there, in a phase the research names after the highest-paid person's opinion: evidence is nothing in the face of it. So the product is not a feedback collector with better summaries. It is an instrument for making a call you can survive being challenged on.
Nineteen customers, in their own words.
The research ran in seven phases on live product pages and pricing, June 2026: competitors in three groups, a benchmark of tools outside the category that people already trust with synthesis, a Lean UX canvas, a funnel model, and the UX patterns. Every claim in it links to its source, and two market facts that turned out to be wrong, an acquisition and a pricing change, were corrected in the file rather than quietly dropped.
The journey is built on 19 verbatim quotes from ten public pages, product manager forums and reviews, with screenshots kept as evidence. Every phase stands on a real quote, and the two phases that rest on a single strong voice are marked as the weaker ones. The low point, defending a call under challenge, scores minus five; the journey the product is designed for lifts it to plus four.
Sceptical users trust what they can open.
The finding that reshaped the design was about trust: this user does not believe AI synthesis, and will not trust a tool that gives a confident answer without showing its work. That is awkward for a product whose engine is AI synthesis. It means the summary is not the value. The summary is the claim, and the product only earns anything if the claim can be opened, inspected and challenged, live, in the meeting.
It does not win by summarising better. It wins by being challengeable.
Specified before it was drawn.
Seven entities, ten clusters and 39 pages, each with a written specification of its blocks, states and search metadata, before any wireframe existed. Synthesis is the home at tap zero, the raw evidence is two taps away, and building a brief to share is one. Three navigation entries instead of five, because a tool that argues for calm under density cannot open with a wall of tabs.
The grey prototype renders that structure and nothing else: 17 screens and 38 pages, every state its own page, with eleven critique findings found and fixed. The visual layer is in progress; this case shows the part that is finished.
Four calls, and what each one cost.
The evidence count ships on day one, not in version two.
Every theme shows how many items it rests on, with a low-signal badge. A single confidence score was rejected because a score hides the count, and the count is what the sceptical user checks.
CostA busier row than a clean ranked list.
CSV first, then Intercom.
CSV lets a product manager work with their own data on day one without asking anyone for admin access. Zendesk and call recordings were deferred because they would delay the first real synthesis.
CostA weaker integrations page at launch than the category's.
Product managers only, for now.
Closing the loop with the customer is a real job for support teams. It has no first-release screen and sits in the trace matrix as a deferral with a reason.
CostA second buyer inside the same company is left for later.
Unsourced numbers are removed, not rounded.
The time a manual synthesis takes is widely quoted and poorly sourced. The research keeps the order of magnitude, hours, and marks the exact figure as open until interviews measure it.
CostA softer pitch than a precise “saves 4 hours a week”.
A structure that holds, and the bets that need people.
Shipped
Built to move
What this file does not know yet.
[?] Does transparency actually buy trust. The whole product is a bet that a product manager will trust a synthesis they can open and challenge. It closes in a prototype test with real product managers and their own data.
[?] How much time it saves. The order of magnitude is defensible, the exact figure is not yet, and it stays an open question rather than a marketing number.
Brio