Sheldon Wang

Sheldon Wang

Seasoned iOS Veteran, Mentor

Design a Push Notification Personalization System

Published:

🎯 Problem

Build a service that sends personalized notifications at the right time and frequency.

Goals:

  • increase open/engagement rates
  • avoid spam and fatigue
  • honor user preferences and compliance rules

📐 Architecture Sketch

flowchart TD; Events["User Events"]-->FeatureStore["Feature Store"]; FeatureStore-->Ranker["Notification Ranker"]; Campaign["Campaign Rules"]-->Ranker; Ranker-->Policy["Policy & Frequency Guard"]; Policy-->Scheduler["Send-Time Scheduler"]; Scheduler-->Provider["APNs/FCM Gateway"]; Provider-->App["Mobile App"];

🔑 Decision Points

Rule engine vs ML ranking

  • Start with rule+score hybrid for explainability.
  • Move to learned ranking when data quality is strong.

Frequency capping

  • Global cap (e.g., max 3/day)
  • Per-category cap (promo vs transactional)
  • Cooldown after dismissals

Quiet hours and timezone correctness

  • Store timezone per user
  • Schedule send windows locally
  • Fall back safely when timezone unknown

🧪 Scoring Inputs

  • recency/frequency of relevant activity
  • previous open/click behavior
  • content category affinity
  • predicted fatigue signal

🚨 Failure/Safety Controls

  • provider outage -> retry queue with dedupe key
  • malformed payload -> validation gate before send
  • runaway campaign -> emergency campaign kill switch

📊 Metrics

  • delivery rate
  • open rate / click-through
  • unsubscribe and disable-notification rate
  • complaint rate / negative feedback

✅ Interview Tip

Say explicitly: optimizing opens alone can hurt long-term retention; include negative-signal metrics in objective function.