~/learn/Product Metrics · Diagnosing a metric drop

Case study: Reels views drop 6%

A full worked answer to a realistic product case, step by step.

7 min readlesson 12 of 15case studyreelswalkthrough

why this matters

Walking through a full case end to end is exactly what a product analytics round looks like.

Prompt: 'Instagram Reels views dropped 6% week over week. Walk me through how you'd investigate.' Here's how a strong answer flows.

1 · Clarify. Is it total views or views per user? Measured how — plays over 3 seconds? Global? Sudden on one day or gradual? Suppose: total views, global, started abruptly on Tuesday.

2 · Validate. Check whether a logging change or new app version altered how a 'view' is counted, and whether the data pipeline is complete. Suppose the data is clean.

3 · Decompose. Total views = Reels viewers × views per viewer. Suppose viewers are flat but views per viewer fell 6% — so people still come, they just watch fewer Reels.

4 · Segment. Split by platform, app version, country and new vs existing users. Suppose the drop is concentrated on iOS, on the newest app version released Tuesday.

5 · Hypothesise. The release could have slowed video loading, changed the feed ranking, or made the Reels tab harder to reach. Checking load-time data shows iOS start-up time for videos doubled.

6 · Check the ecosystem. Did Feed or Stories views rise (cannibalization) or did total time on app drop? Suppose total time also fell — real lost engagement, not a shift.

Answer outline

Clarify → Validate → Decompose (viewers × views/viewer)
→ Segment (iOS, new version) → Hypothesis (slow loading)
→ Ecosystem check → Recommend: fix/rollback, add alert

Pro tip

Recommend: roll back or hotfix the iOS video player, add a load-time guardrail alert, and re-measure after the fix. Sizing the impact ('about 6% of Reels views, mostly iOS') makes the answer feel senior.

Interview trap

Don't stop at 'it was the iOS release'. Interviewers want the follow-through: confirm the mechanism, check the wider product and recommend a concrete action.

key takeaways

  • Decompose into viewers × views per viewer.
  • Segment to localize (platform, version).
  • Check the ecosystem, then recommend.

check your understanding

Reels viewers are flat but views per viewer fell. What does that tell you?