🌏 中文版
Today's Focus
A product sense interview has never really been about whether you know a framework — AI lets anyone generate a structurally complete answer in seconds, so structure alone is now just the floor. What actually decides the score is the five dimensions that Lenny's Newsletter and former Meta PM Ben Erez laid out in 2025, and they're graded independently, not as a weighted average: a beautifully chosen segment doesn't rescue a problem-identification step that's really the same pain stated three different ways. Today's drill applies that five-dimension framework to a real 2026 Meta phone-screen question.
Framework Cheat Sheet
Ben Erez's five-dimension framework (Lenny's Newsletter, April 2025)
This framework ties "what to say" directly to "how you're scored," with a suggested time budget for each step:
| Dimension | What you do | Time | Common failure |
|---|---|---|---|
| 1. Clear communication | State your plan and assumptions up front so the interviewer always knows where you are | 3–5 min | Stating the structure once at the start, then drifting without signposting |
| 2. Product motivation | One sentence on why this product should exist, tied to the company's mission | 3–5 min | Skipping straight from clarification to segmentation |
| 3. Segmentation | Map the ecosystem, name 2–4 segments, pick one and justify it | 8–10 min | Generic buckets like "light / regular / power users" that could describe any product |
| 4. Problem identification | Walk through that user's real moment of use, list pains, pick one | 8–10 min | Three "pains" that are really one emotion restated three ways |
| 5. Solution development | Brainstorm, rank by impact vs. effort, define version one | 8–10 min | Three "solutions" that are the same idea wearing three different UIs |
Lenny's Newsletter is explicit on this point: "excellence in one area can't compensate for weakness in another" — the same discipline as last week's drill about solutions that can fail independently, except here even the scoring sheet keeps each dimension separate. IGotAnOffer, citing a former Meta product lead, calls segmentation the interview's "lynchpin": once the segment is generic, the problem and the solution inherit that genericness automatically.
Today's Practice Question
The question
"You're a PM at Meta, assigned to design a brand-new product that connects home-repair workers — electricians, painters, plumbers — with consumers. How would you approach it?" (Source: real Meta PM phone-screen question, reported by a 2026 candidate, collected by Aced/Exponent)
How to break it down
- Clarify the problem: Start with questions about the question itself — is this a standalone app, or a feature inside Facebook/Marketplace? Nationwide, or one city first? These answers directly determine whether you can lean on Meta's existing social graph and user habits, so they're worth asking before anything else.
- Product motivation: Tie it to Meta's mission — not "help people find a worker," but "extend the trust people already have in each other into the home-repair moment," which lines up with Meta's own framing of "connecting people."
- Pick a segment: Home-repair users roughly split into "first-time homeowners who have no idea who to call," "longtime homeowners who already have a go-to crew but just can't reach them right now," and "landlords/property managers juggling multiple units." Pick the first group — their pain is the most distinctive, and it's the one Meta's existing neighborhood-group data can actually help with.
- Find the pain point: Walk through that user's real day. A pipe leaks; they search online and get conflicting answers about whether it's a plumber or an electrician's job; they ask friends and get three different opinions; they open an existing home-services app that forces them to pick a service category first, and they don't even know which category applies. Run it through the four buckets — time, money, motivation, trust and safety — and the root cause here is "I don't know which trade to call," which is a genuinely different problem from "I found someone but can't tell if the quote is fair."
- Generate solutions: Use "product cuts" to check that your options are actually distinct — (a) AI symptom intake: type, photograph, or describe the problem by voice, and let the system infer the right trade; (b) pull in Meta's existing neighborhood-group posts to surface "people nearby have recommended this crew"; (c) a moonshot: fold quoting, booking, and payment entirely into a Messenger thread. Each option bets on a different lever — AI judgment, social trust data, or a closed transaction loop — so each one can fail on its own.
- Define success: The north star could be "share of symptom descriptions that convert into a completed booking"; the input metric is the completion rate of the AI symptom-intake flow; the counter-metric is the rate of mismatched-trade complaints — which matters especially because it's exactly the risk Thumbtack itself is managing in today's case.
Sample answer (how this could actually sound in the room)
Lead with motivation and segmentation: "I'd frame this as extending the trust people already have in their Meta network into the moment they're most anxious about: home repair. I'd anchor on first-time homeowners with little or no repair experience, because their problem isn't 'I can't find a worker' — it's that they don't even know which type of worker to look for. That's different from someone who's lived in the same house for years and already has a crew."
Move into pain and solution: "What this group actually describes is a symptom, not a service category — a leak could be plumbing, roofing, or something else entirely. So version one would be an AI symptom intake: let people describe what's happening in text, a photo, or voice, and have the system infer which trade is needed, instead of forcing them to pick a category on entry. I'd also pull in Meta's existing neighborhood-group data so the match shows 'people near you have recommended this crew' — that's something a plain marketplace can't offer."
Close on metrics: "I'd use 'share of symptom descriptions that convert into a completed booking' as the north star, and watch the mismatched-trade complaint rate as a counter-metric closely, because if the AI guesses the wrong trade, trust ends up worse than if the feature never existed."
Self-check
| Check | Covered? |
|---|---|
| Does product motivation tie back to the company mission, not just describe the feature? | |
| Is the segment specific enough that it could only belong to this product, not a generic bucket? | |
| Are the pain points genuinely distinct root causes, not the same emotion restated three times? | |
| Can each solution fail independently, rather than being the same idea in a different UI? | |
| Did you name a north star, an input metric, and a counter-metric? | |
| Bonus: did you proactively name a new risk your own solution could introduce (e.g., mismatched trades)? |
Today's Case
Thumbtack: replacing category-first search with AI-guided symptom intake
On April 30, 2026, Thumbtack formally announced it was replacing its core discovery flow — picking a service category first — with one where users describe a home problem in text, photo, or voice, and AI infers which type of professional they need. The company cited its own survey: more than half of homeowners admit they misdiagnose their home problems, and nearly 85% aren't confident about which kind of professional to call — which is the real-world version of today's practice question's "I don't know which trade to call" pain point, not something an interviewer invented. In early testing, 87% of users said being able to express their problem through a photo or voice felt valuable, and users on the new flow reported more confidence moving forward than those using traditional search. CEO Marco Zappacosta put it plainly: "In an industry where word-of-mouth has long guided how homeowners find help, AI gives us an opportunity to recreate that confidence at scale." Senior Director of Product Waqas Sheikh added the line that maps most directly onto today's framework: "Home projects don't always start as perfect plans — they often start as questions, problems, or ideas. With our new AI-powered experience, we've designed for the way projects really begin."
Interview angle: this case is direct evidence for the "AI symptom intake" step of today's practice answer — when explaining why you wouldn't force users to pick a category up front, you can cite Thumbtack's own survey numbers instead of a vague claim that "users don't know what to pick." It also answers a likely follow-up: how would you know if the AI-trade-guessing solution is working? Thumbtack's answer was to compare confidence and downstream conversion between users on the new flow and users on traditional search — exactly the kind of comparison today's practice question asks for in step 6.
Further Reading
- Product Sense Interview: Framework, Examples, and Practice — full source of today's framework, with a worked example on improving Google Maps
- How to Crack AI Product Sense Interviews — covers the "Human Delta" concept for the AI era: what judgment you add on top of what AI can already produce
- The Definitive Guide to Mastering Product Sense Interviews — the original Lenny Rachitsky / Ben Erez piece that today's framework and scoring dimensions are drawn from
References
- Thumbtack Introduces AI-Powered Experience to Reinvent How Homeowners Care for Their Homes (official press release, 2026-04-30) — source of the survey figures, early-testing results, and executive quotes used in today's case
- Product Sense Interview: Framework, Examples, and Practice — source of the five-dimension scoring framework and time budgets
- 52 Real Product Manager Interview Questions (2026 Guide) — source of the practice question (Meta's home-repair product prompt)