· PM Editorial · Product Sense · 5 min read
Improve Spotify Discover Weekly: Metrics and North Star
How to pick a north star metric and supporting KPIs for Spotify's Discover Weekly, covering save rate, skip rate, completion rate, and a diversity index.
Interviewers frequently follow up a Discover Weekly product sense question with “okay, how would you measure success?” This is a distinct skill from generating feature ideas — it tests whether you can build a metrics tree that actually captures the tradeoffs your proposed features create. This post walks through choosing a north star metric and the supporting metrics tree for Discover Weekly.
Why Metrics Questions Trip Candidates Up
Candidates who did well on the feature-brainstorming half of the question often stumble here because they reach for a single vanity metric — “I’d measure how many people listen to the playlist” — without acknowledging that Discover Weekly has at least three plausible business goals that pull against each other: engagement (keep people listening), discovery (expose new artists), and retention (justify the subscription). A good metrics answer names the tension and picks a north star deliberately, rather than picking the first number that comes to mind.
Step 1: Separate Business Goal From Product Goal
The business goal behind Discover Weekly is subscriber retention and increased listening hours, both of which correlate with lower churn and higher ad/subscription revenue. The product goal — the thing the feature is directly built to do — is surfacing relevant, novel music each week. A good north star sits at the intersection: it should move when the product goal is met, and moving it should plausibly cause the business goal to improve.
Step 2: Candidate North Star Metrics
| Candidate metric | Definition | Pros | Cons |
|---|---|---|---|
| Playlist save rate | % of the 30 tracks saved to a user’s library | Strong explicit intent signal | Some users never save, even when they love a song |
| Skip rate | % of tracks skipped within first 30 seconds | Cheap, high-volume signal | Ambiguous — skip could mean “wrong mood,” not “bad song” |
| Completion rate | % of tracks played to at least 80% duration | Captures passive engagement, not just active curation | Doesn’t distinguish liking a song from just not bothering to skip |
| Diversity index | Spread of genres/artists relative to user’s existing library | Directly measures discovery, the feature’s stated purpose | Can be gamed by injecting random tracks that hurt relevance |
Step 3: Choose the North Star
The strongest answer names playlist save rate, adjusted for weekly active listeners, as the north star — with the reasoning stated explicitly: saving a song to your library is the highest-intent action a user can take on a recommendation, it’s a leading indicator of both engagement (people build libraries they keep listening to) and discovery (a saved song from an unfamiliar artist is a genuine discovery win), and unlike completion rate it can’t be satisfied by passive non-action.
State the caveat too: save rate under-counts users who love a song but forget to save it, and it slightly under-weights genuinely serendipitous discovery, since surprising tracks get saved at a lower initial rate even when they eventually become favorites. That’s why it needs supporting metrics, not to stand alone.
Step 4: Build the Supporting Metrics Tree
Tier 1 — North star Weekly playlist save rate (saves per active listener per week)
Tier 2 — Diagnostic drivers
- Skip rate within first 30 seconds (inverse leading indicator — high skip rate predicts low save rate)
- Full-playlist completion rate (proxy for passive satisfaction)
- New-artist diversity index (are we actually diversifying, or just recommending safe, familiar-adjacent tracks)
- Repeat-listen rate 30 days post-playlist (do saved songs actually get replayed, or saved and forgotten)
Tier 3 — Guardrails
- Overall app session length (make sure changes to Discover Weekly don’t cannibalize other listening)
- Unsubscribe/skip-the-feature rate (are users opting out of Discover Weekly entirely)
- Cold-start segment save rate specifically (make sure aggregate numbers aren’t masking a struggling new-user segment)
Step 5: Map Metrics to Segments
A north star that looks flat in aggregate can hide diverging segment trends, so a rigorous answer breaks the tree out by the same segments used in the feature-ideation half of this question:
| Segment | Best-fit metric | Why |
|---|---|---|
| Power discoverers | New-artist diversity index | Already engaged; the marginal value is exposure to more unfamiliar artists |
| Passive listeners | Completion rate | Rarely take explicit actions; passive listening is the honest signal |
| Genre loyalists | Skip rate + save rate together | Watch for skip rate spikes if diversity floor pushes too far outside their lane |
| Cold-start users | Day-30 save rate and retention | Cold start success is retention-shaped, not discovery-shaped, in month one |
Step 6: Connect Metrics Back to the Proposed Features
If you proposed a “discovery dial” (more familiar / more adventurous) in the feature half of the question, the natural metric pairing is save rate segmented by dial setting — if adventurous-mode users show a lower save rate but higher new-artist diversity index and stable 30-day retention, that’s evidence the feature is working as designed, not failing. This is the kind of connective tissue interviewers are listening for: metrics that were clearly chosen because they validate a specific hypothesis, not metrics chosen because they’re commonly cited.
A Simple Test for Any North Star You Propose
Before stating your north star out loud in an interview, run it through three questions: does it move when the feature succeeds, could it be gamed in an obviously bad way (like injecting more tracks to inflate completion rate), and does moving it plausibly cause the actual business outcome to improve. Playlist save rate passes all three for Discover Weekly; skip rate alone fails the second and third.
Takeaway
The strongest metrics answers do three things: name the tension between competing goals explicitly, pick one north star with a stated reason instead of listing five metrics with equal weight, and build a small supporting tree of diagnostics and guardrails that would actually catch the north star being gamed. Practice this pattern on Discover Weekly and it transfers directly to any recommendation or feed product — Netflix homepage, TikTok For You, LinkedIn feed.
For more worked metrics-tree examples across marketplace, growth, and consumer social products, see The 100x Product Manager Interview Playbook (Amazon: https://www.amazon.com/dp/B0DBC1FQWH?tag=sirjohnnymai-20).