Role (simulated)
Product Manager: Retention & Engagement
Goal
Fix the home screen signal: reduce churn
North Star
+15–20% home screen session starts
Watch time proxy
+10–14% hrs/subscriber/month
Guardrail
~0 rise in "lost progress" tickets
Netflix spends billions on content:
A cluttered row stops subscribers from ever reaching it.
Every session that ends in "I couldn't find anything to watch" is a data point toward cancellation. The Continue Watching row: full of shows watched 3% once in 2021: is poisoning recommendation signals and pushing the subscribers most at risk of churning toward the exit.
1

Content Strategy Layer: The Retention Loop That's Breaking

Netflix's business model is a content investment loop. When Continue Watching corrupts that loop, it doesn't just hurt UX: it undermines the entire content strategy machine.

Netflix's Content: Retention Loop
Where Cleanup Mode fits and why it matters beyond the UI
💰
Content Investment
Netflix spends ~£2.5bn/year on UK content. Content is the product.
😵
Discovery: BROKEN
Continue Watching cluttered. 3:1 impression gap. Users scroll past. Content investment wasted.
📊
Watch Signal
Engagement data feeds the algorithm. Corrupt data: bad recs: worse discovery.
🔄
Retention
Hours watched/subscriber is the primary churn predictor. Low hours: cancel.
🧹
Cleanup Mode: FIX
Clean row: clear discovery: better signals: stronger recs: higher retention.

This is a product strategy argument, not just a UX argument. Netflix's recommendation engine is only as good as the intent signals it receives. A row full of shows watched 3% three years ago sends false "in-progress" signals. Cleanup Mode doesn't just clean the UI: it cleans the data. That's a recommendation quality improvement and a competitive advantage.

2

User Segmentation: Who's Most at Risk

The Binge-then-Quit
High volume: Low loyalty: ⚠️ Highest churn risk
Tries 4–6 shows/month, rarely finishes. Row fills fastest. When the UI overwhelms them, they close the app and eventually cancel. Design Cleanup Mode's speed for this segment first.
Est. ~35%: Primary target
The Loyal Re-engager
Medium volume: High loyalty: Low churn risk
3–4 active shows in progress. Archive is their specific feature: they genuinely want to "park" shows they'll return to. Progress preservation matters most here.
Est. ~45%: Benefits from Archive
The Casual Browser
Low volume: Medium loyalty: Medium churn risk
Opens Netflix 1–2x/week to "see what's on." Scrolls past Continue Watching entirely: Recommendations never gets seen. Cleanup restores the discovery surface they're looking for.
Est. ~20%: Benefits from AI nudge
3

Why This Problem First

Business impact
Continue Watching is on every home screen, every session. Fixing its impression-to-start ratio has platform-wide reach. Massive leverage.
Churn signal
"Couldn't find anything" is the #2 cancellation reason. This has a direct, researched connection to revenue loss: not a soft satisfaction problem.
Competitive urgency
Disney+ and Apple TV+ have smaller libraries: their rows are naturally shorter. Netflix's scale is creating a UX debt competitors don't have. The longer we wait, the larger the gap.
4

Problem Quantification

3:1
Impression-to-session-start ratio for Continue Watching vs ~1.4:1 for Trending Now. A 2× gap on the most prominent row.
Hypothesis from engagement research
17+
Average shows in a UK subscriber's Continue Watching row: most from "let me try this" clicks that never converted to real engagement. Noise drowning signal.
Estimated from user research
£216
Annual value of one retained premium subscriber. One subscriber saved from "I couldn't find anything" churn = £216 retained. The ROI maths is simple.
Based on £17.99/month UK premium tier
5

Hypothesis

🧪
Core Hypothesis

If subscribers can triage their Continue Watching row in under 60 seconds: with an archive that preserves progress: then home screen session starts will increase 15–20%, Recommended For You CTR will increase ~25%, and watch time per subscriber will increase 10–14%.

The design insight: Users don't want to delete progress: they want to defer it. Archive is not a rebranded delete. It's a genuinely different action: hide from the row, preserve progress forever, restore in one tap. This distinction is what makes the feature behaviorally safe enough that users will actually use it at scale. Any version without Archive will be behaviorally abandoned.

6

The Solution: Cleanup Mode

🚧 Root Causes
  • Every watch: even 2% of a pilot: enters the row with no expiry
  • Removal is one-by-one, buried in a long-press menu users never find
  • Remove = lose progress forever: users fear this so they do nothing
  • Algorithm treats "watched 3% in 2021" as in-progress, corrupting recommendations
✦ Cleanup Mode
  • Entry: long-press on Continue Watching row, or AI nudge after 90 days idle
  • Archive: hides from row, preserves progress forever, one-tap restorable
  • Remove: clears entry + progress for truly abandoned shows
  • Swipe gestures for speed; tap buttons as safe fallback; undo always available
Continue Watching
Stranger Things
Wednesday
Peaky Blinders
← 14 more: 3:1 impression gap
17 shows. Decision fatigue: churn risk.
Before: loop broken
🧹 CLEANUP MODE
✨ Smart Suggestion
Peaky Blinders: idle 4 months (S1E1: 5%). Archive?
📦 Archive
✕ Remove
Keep
← swipe to archive: → to keep
Stranger Things
S4E1: 12%
✓
📦
Wednesday
S1E8: 88%
✓
📦
Peaky Blinders
Archived: progress saved
✓ Signal clean. Recs improving.
Cleanup Mode
✓ Loop restored. Sessions ↑
Continue Watching: 4 shows
Stranger Things
Wednesday
Adolescence
Recommended For You
Black Mirror
The Crown
Baby Reindeer
After: loop fixed
7

Tradeoffs: The Decisions That Shaped This

DecisionRejectedChosenThe honest tradeoff
Remove vs ArchiveDelete forever✓ ArchiveDelete-all is the obvious answer and the wrong one. Fear of losing progress is the reason users don't clean up. Archive is the only design that changes behaviour at scale. Cost: complex data model, GDPR retention policy needed. Without Archive, no version of this feature works.
MVP triggerAI nudge only✓ Edit button (v1) + AI nudge (v2)AI nudge adds ML complexity. Ship the Edit button first to validate demand. If power users who find it don't clean up, the AI nudge won't save a fundamentally broken feature. Validate before investing in discoverability.
User control vs algorithmAuto-sort the row✓ User controls their rowAuto-sort removes agency: the same force that created the problem. User-controlled cleanup also generates explicit intent signals that improve the recommendation engine as a secondary win.
Row count visibility"17 shows" badge on home screen✓ Count only inside Cleanup ModeOn the home screen, "17 shows" creates passive anxiety that suppresses session starts. Inside Cleanup Mode, the same number creates active motivation. Context determines psychology. Passive display is a dark pattern.
8

What I Would NOT Build in v1

🚫 "Delete all" button

Would generate a wave of "I lost my progress" support tickets. Archive solves the same job without the irreversibility. Delete-all is never the right PM call when user data is involved.

🚫 Algorithmic row re-sort

Removes user agency: the opposite of what we're restoring. Requires extensive testing to avoid re-surfacing archived content. Complex, risky, and lower value than giving users direct control.

🚫 TV remote triage in v1

D-pad interaction requires a completely separate design pattern. Mobile is the highest-frequency churn surface. Launch there first, validate the feature works, then invest in TV as a separate workstream.

9

Execution Reality

Sprint Plan: 2 engineers (iOS + Android) + 1 designer + PM + 1 data analyst
Sprint 1 (2 weeks)
Archive data model + Edit button entry. Ship to 5% cohort. Track: does anyone use it? What's the session start delta post-cleanup?
Sprint 2 (2 weeks)
Swipe gestures + AI idle detection. Only if v1 shows meaningful cleanup adoption: don't add ML complexity before validating demand.
Critical risk
Archive data model must guarantee zero accidental progress loss. Any failure destroys trust and requires emergency rollback. Extensive edge case testing is non-negotiable before ship.
Data Science / ML
Signal quality improvement
90-day idle threshold is a hypothesis: validate by genre. Archive actions generate new explicit intent data that could improve the recommendation engine. Worth surfacing to the data team early.
Product Analytics
Experiment design risk
Users who clean up are inherently different from those who don't: selection bias in a simple A/B test. Need pre/post within-user measurement. Get the data team involved before Sprint 1.
Legal / Privacy
GDPR data retention
Archived items can't be stored indefinitely without a defined retention policy under UK GDPR. "Forever" is a product promise but needs legal definition. Confirm before launch: not as a post-launch amendment.
10

Metrics: Connected to Subscription Revenue

North Star
+15–20%
Session starts from home screen for Cleanup users vs matched control. Maps to hours/subscriber: Netflix's primary churn predictor.
If flat: decision fatigue isn't the root cause of the session gap. Investigate the row content or algorithm next.
Guardrail
~0 rise
In "I lost my progress" support tickets. Any increase = Archive is failing. This is an immediate rollback signal. Subscriber trust > feature.
Monitor weekly in first 4 weeks. Non-negotiable.
Leading
+20–28%
CTR on Recommended For You row post-cleanup. Freed attention redirects to discovery. Also validates signal quality hypothesis.
Visible within 2–4 weeks of first cleanup session.
Case Study 02 of 03: Retention Loops: Content Strategy
← Uber EatsNext: CMS Task Routing →