Product Builder Interview Drill — 2026-10-16: Growth & Experimentation

🌏 中文版

Today's Focus

Friday is Growth & Experimentation. The last few days covered whether a feature is worth building and how to define success; today is a sharper problem: a number drops, or your manager hands you "figure out how this grows itself," and in 30 minutes you have to turn it into a funnel, a diagnosis, a ranked experiment and a guardrail metric that holds up under questioning. Growth interviews punish candidates who skip the diagnosis and jump straight to a prescription — add a share feature, add a push notification. What the interviewer is actually watching for is whether you find the leak first.

Framework Cheat Sheet

AARRR: split the funnel into five stages first

Dave McClure's 2007 "pirate metrics" split growth into five stages you can diagnose separately:

Most candidates can recite the table; what interviewers actually want to hear is whether you ask "which stage does 'growth' mean here" before reciting all five.

A loop isn't a funnel: switch frameworks for strategy questions

Reforge's Brian Balfour argues that a funnel splits each stage across different teams — marketing owns acquisition, product owns retention, sales (in B2B) owns revenue — and that split misses how a product actually grows itself. A loop is different: one cycle's output becomes the next cycle's input, so it compounds instead of just converting. When asked a strategy question like "how would this product unlock more growth," answering with a funnel gets pushed back on. Answer with a loop instead:

1. Name the loop that already exists (a user invites teammates, teammates create content, content pulls in more users)
2. Find the weakest link in that loop (invited users who never post, say)
3. Propose one or two new loops
4. Measure loop health by cycle time and how many new users each cycle produces

Ranking experiments: ICE

List three or four candidate experiments, then score each on Impact × Confidence × Ease. Sean Ellis popularized the shortcut; Itamar Gilad's addition is the part that matters: the confidence score should come from evidence, not optimism — "I have a feeling this will work," with no data or prior case behind it, doesn't earn a high score.

Today's Practice Question

The Question

Suno's mobile app has strong first-song generation, but only 18% of new creators make a second song within 7 days. How would you diagnose the drop-off, break down the funnel, pick the most important leading and lagging metrics, and prioritize the first round of product changes or experiments?

(Source: a Growth PM interview question from the AllthingsPM question bank)

Breaking It Down

1. Clarify: confirm what "growth" means here. The question already frames "making a second song" as the retention signal, so assume the goal is building a weekly-creation habit, not revenue.
2. Map the funnel: install → first song → listen/share → return visit → second song. The drop happens between the first song and the second.
3. Structure the analysis: segment the 82% who didn't come back — did they share their first song? Did they hit a free-credit limit? Did they arrive through a shared link or through search? Someone who came in to listen to a friend's song has very different intent from someone who searched "AI music generator," and averaging the two hides the real problem.
4. Propose a solution: candidate experiments — a remix prompt on the result screen (regenerate the same song in a different style), a day-two notification with a style suggestion, or saving drafts so users can come back and finish. Rank with ICE; the remix prompt is usually cheapest and sits closest to the moment the user just felt the value.
5. Define success: the primary metric is the share of new creators who make a second song within 7 days; the leading metric is day-one remix taps, which signals direction before day 7; the guardrail is 30-day retention and compute cost per user, so you don't trade free credits for a pile of songs nobody comes back to make.

Sample Answer (how you'd actually say this in an interview)

Break up the 18% before reacting. I wouldn't jump to "add a share button," because the 82% who didn't make a second song probably aren't one group — some ran out of free credits, some listened to a friend's shared song with no intention of making their own, some generated a song, weren't happy with it, and just left. I'd pull the last two weeks of new creators and segment by whether they shared their first song, whether they arrived through search versus a shared link, and whether they hit the free-credit limit, then see how much of the 82% falls into each bucket.

Pick the cheapest experiment closest to the value moment. Say the segmentation shows the biggest bucket is people who generate a song and leave without being nudged toward a second one. I'd test a remix prompt on the result screen first — regenerate the same song in a different style — because it's a small, low-cost change that sits right after the moment the user just felt "I made this." A day-two notification and draft-saving come later in the queue, because I'd want the remix results back first rather than guessing.

Watch three layers of metrics, not one number. The primary metric is the share of new creators making a second song within 7 days; the leading metric is day-one remix click rate, which tells me the direction is right before I wait a full week; the guardrail is 30-day retention and compute cost per user — if the remix prompt lifts the second-song rate but those users are gone by day 30 anyway, it means I just traded free credits for a pile of short-lived creations, not a real habit.

Self-Check List

Use this table to check your answer for gaps:

Today's Case

Calendly: scheduling is a problem both sides hit at the same moment

Blake Bartlett, a partner at OpenView and an investor and board member at Calendly, explained in a Sacra interview how Calendly grew a loop: scheduling isn't a single-player task — the person who sends the link and the person who receives it are both stuck on "we need to pick a time" at the same moment. The recipient clicks the link, sees their own time slots to fill in, and when it's their turn to schedule someone else, they sign up for an account and send out their own link. Using the product is itself the act of promoting it.

Interview angle: when asked how to judge whether a product is worth investing in a viral loop, this case is the test, not a template to copy. The point isn't "add an invite feature" — it's "the core workflow already involves someone who isn't a user yet, and that person has the exact same need at the exact same moment." Most B2B products are single-player; bolting on a share prompt or a referral reward can't manufacture that condition, and it only adds friction to a flow that was working fine.

Further Reading

• Growth PM Interview Questions: The Framework, Worked Examples and a Practice Plan (2026) — the full five-step framework with three worked real questions
• Growth Loops Explained: What They Are, Types & How They Beat Funnels — four types of loops with examples
• 7 Product-Led Growth Examples for B2B Companies — seven PLG motions, what each needs to be true, and where each breaks

References

• Growth PM Interview Questions — AllthingsPM — framework: the five-step answer spine, AARRR, loops vs. funnels, ICE; source of the practice question (Suno's retention drop-off)
• Growth Loops Explained — Growth Method — loop definition and the four types
• 7 Product-Led Growth Examples for B2B Companies — The Higher Pitch — today's case: how Calendly's viral loop works
• Blake Bartlett, partner at OpenView, on the future of product-led growth — Sacra — original interview source for the Calendly case