Systems Atlas

Chapter 1.3

Search vs Discovery vs Recommendation

Three different problems that are often confused. Understanding the difference is crucial.


Search

User knows what they want.

  • Input: Explicit query
  • Output: Ranked list
  • Goal: Precision
  • Intent: High
Example: "running shoes size 10"

Recommendation

System suggests what user might like.

  • Input: User history
  • Output: Personalized suggestions
  • Goal: Engagement
  • Intent: Low
Example: "Because you watched..."

Discovery

User is exploring, not hunting.

  • Input: Clicks, filters
  • Output: Navigation UX
  • Goal: Exploration
  • Intent: Medium
Example: Browse Electronics → Phones

Side-by-Side Comparison

AspectSearchRecommendationDiscovery
User IntentHigh (knows what)Low (open to ideas)Medium (knows category)
PersonalizationLow-MediumVery HighMedium
Latency SensitivityVery HighMediumLow
Success MetricCTR, ZRREngagement, CVRBrowse depth

When to Use Which

Search: The Direct Path

  • • Users with specific intent
  • • Transactional queries ("buy X")
  • • Repeat purchases
  • • Research queries ("compare X vs Y")

Recs: The Suggestion Engine

  • • Homepage personalization
  • • "You might also like" sections
  • • Email campaigns
  • • Cart upsells

Discovery: The Exploration Flow

  • • New users exploring
  • • Window shoppers
  • • Category exploration
  • • Gift shoppers

The Hybrid Reality: Netflix

Netflix combines all three seamlessly:

Search

User types "comedy"

Discovery

Browse by genre, trending

Recommendation

"Because you watched..."

Common Mistakes

"We have Elasticsearch, search is done"

Reality: Search needs ranking, personalization, and continuous improvement.

Over-Personalizing Search

Reality: "iPhone" should show iPhones, regardless of who searches.

Ignoring Discovery

Reality: New users don't know your catalog. Discovery helps them explore.

Separate Teams, No Coordination

Result: Conflicting experiences between search and recommendations.

Key Takeaways

01

Search

High intent, precision-focused (User: "I want X")

02

Recommendation

Low intent, engagement-focused (System: "You might like Y")

03

Discovery

Medium intent, exploration-focused (User: "Show me category Z")

04

Hybrid Reality

The best products (e.g., Netflix, Amazon) combine all three seamlessly.