User Research on Drop-offs and Content Relevance in Inshorts
Two Inshorts interviews that confirm some drop-off hypotheses and cleanly kill others
Phase 01 · User research
User Research on Drop-offs and Content Relevance in Inshorts
✳The problem they went after
Inshorts sells quick current-affairs reading to Indian working professionals, and those readers keep leaving mid-session. This phase goes after why, and which problem deserves P0 status rather than treating every complaint as equally urgent. The starting hypotheses cover content relevance, notification volume, ads, and the inability to resume where reading stopped.
✳What they actually did
Two interviews, read closely rather than counted. A casual reader, previously using the app for group discussion prep, confirmed friction on relevance, notification volume, lost reading progress, and appetite for hyperlocal news. A content-savvy reader confirmed the save-progress gap and the hyperlocal interest but contradicted the relevance and notification hypotheses, finding both acceptable. Neither knew Videos or Insights existed inside the app. Ads, loudly criticised in store reviews, did not register as a problem for either.
✳What you can take from it
Worth stealing
- Disconfirming interview evidence reported, not buried
- Store review complaints tested against real users
- Feature awareness gap surfaced without being asked about
- Competing news sources named: Instagram and LinkedIn
Read this if
Learning to run interviews capable of disproving the hypotheses behind them
The transferable bitA contradiction is kept rather than smoothed over. One reader disagrees with two hypotheses, the write-up says so, and priorities are then ranked on what both users agreed about: repetitive content and no way to resume. Store review complaints being overturned in conversation is a useful reminder about where complaint data comes from.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
Phase 02 · Solution prioritisation
Personalizing Inshorts News Feed to Reduce Drop-offs
✳The problem they went after
Working professionals on Inshorts hit repetitive and irrelevant stories with no lever to pull. There are no granular interest controls and no way to manage sources, so everyone receives the same undifferentiated stream. The result is early drop-off and shorter sessions among exactly the audience the app was built for.
✳What they actually did
Ten solutions generated, then scored on RICE. Three came out ahead: interest-based personalisation where users signal preferences in real time, a source filter and control panel for following or blocking specific publishers, and smart topic clustering that folds related shorts into a single unified story card. The scoring favoured them for hitting the root cause at modest engineering cost, with confidence backed by both interview and survey evidence rather than intuition.
✳What you can take from it
Worth stealing
- Ten ideas generated before any scoring happens
- Top three read as one layered personalisation system
- Topic clustering used to attack repetition directly
- RICE confidence anchored in interview and survey data
Read this if
Prioritising a feed backlog and tempted to pick a winner too early
The transferable bitNotice that the top three are not three competing bets, they are layers: topics, sources and repetition handled separately but designed to work together. That is a stronger prioritisation outcome than crowning one winner, and it only becomes visible because enough ideas were generated for the complementary ones to sit side by side.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
Phase 03 · The PRD
Personalizing the Inshorts Feed for Working Professionals
✳The problem they went after
Professionals abandon Inshorts sessions when the feed repeats itself or drifts away from their interests. There is no topic filtering and no source control, so the only available response is to close the app. Over time that shows up as shorter sessions and sliding thirty-day retention.
✳What they actually did
Three solutions carry the document: interest-based personalisation that learns from in-feed feedback gestures, a source filter panel for following or blocking publishers, and topic clustering that bundles shorts covering the same story into a scrollable card. Targets are stated plainly, at twenty-five percent longer average sessions, fifteen percent uplift in thirty-day retention, and at least forty percent of users actively touching the personalisation controls.
✳What you can take from it
Worth stealing
- Discover Mode as an explicit anti-filter-bubble guardrail
- Auto-suggested sources when blocking shrinks feed volume
- Thirty-day interest reset prevents stale personalisation
- Adoption target set on control usage, not only retention
Read this if
Writing a personalisation PRD that needs guardrails against its own side effects
The transferable bitA Discover Mode counters the filter bubble the personalisation would otherwise build, alternative sources get suggested when blocking thins the feed out, and a thirty-day interest reset stops early signals calcifying. Each guardrail names a specific way the feature could fail on its own terms, which is what makes the set worth copying.
✳The submission itself
Open the PDF →Reading on a phone? The embedded viewer is cramped — open the PDF in a new tab instead.
