Like our product’s briefs, every claim cites its source (W6 card-sort interviews, W7 assumption map, W8 prototype tests), and anything untested goes under Unverified.
Description: What is it?
A tool that turns a PM’s customer feedback, usage data, and engineering estimates into a roadmap decision brief, with every claim linked to its source and weak claims flagged as unsupported, outdated, or conflicting.
Problem: What problem is this solving?
B2B PMs struggle to defend roadmap recommendations to leadership because they can’t quickly tell which of their evidence is strong, outdated, or contradicted. We’re not solving how PMs write the final doc (they already use Notion, W6) or how they set dates.
Why: How do we know this is a real problem and worth solving?
In 12 card-sort sessions (W6), PMs had room for only 6 of 19 features in “Soon.” They kept evidence features: open a claim’s source (12/12), flag weak claims (11/12), and find conflicts (10/12). They cut editing (3/12) and export (2/12). P1: “I don’t need another document generator. I need to know which evidence I can actually defend.”
Success: How do we know if we’ve solved this problem?
4 of 5 new testers finish the workflow unaided in under 5 minutes; 4 of 5 explain a weak claim and find its source; all 5 evaluation checks pass on the AI version (W8); and 1 budget holder agrees to a $20/seat pilot (W7).
Audience: Who are we building for?
B2B PMs who prepare roadmap reviews: enterprise PMs who need traceability and startup PMs who need speed (W6). Their companies are the buyers (W7).
What: Roughly, what does this look like in the product?
Frame the decision → upload materials → review a draft where every claim links to its source and is labeled Supported, Unverified, Contradicted, Dependency, or Status unconfirmed → copy the brief out. After W8, attaching a file will no longer count as verifying a claim.
How: What is the experiment plan?
We test our three riskiest assumptions (W7) on five fictional companies, so no one shares confidential data: the AI flags weak claims (5 evaluation checks), PMs open sources (usability tests with 5 testers), and companies pay $20/seat (a pricing page shown to PMs and budget holders). Our prototype passed only 1 of 4 checks (W8), so we fix evidence labels first.
When: When does it ship and what are the milestones?
Fix v0 labels → connect document analysis → usability and pricing tests by the end of the quarter → pilot in a real roadmap review next quarter.
Business model: Who pays, how, and why?
- Who pays: The PM’s company (W7). PMs are the users who bring it in.
- How: $20 per PM seat per month (W7).
- Why: PMs already write the final version in Notion (P6, W6). We cover the step before that: getting to an answer they can defend. 8 of 12 PMs said they’d pay, and every firm yes named $15 or more (W6).
Unverified: What do we still need to prove?
That PMs check sources in a real review, not just a card sort; that companies, not just PMs, will pay; that companies trust us with internal data (W7); and that AI costs stay well under $20 per seat.
Reading: What idea from this week’s reading did we use, and where?
Lenny says a problem statement should be one focused, solution-free sentence that names what you’re not solving. We used this in our Problem section.
