Systems Atlas

Chapter 1.2

When Search Fails

Revenue, retention, and trust all suffer when search doesn't work.


Real-World Case Studies

Flipkart Big Billion Days

Search latency spiked to 3+ seconds during peak sale. Estimated ₹50Cr+ lost in 4 hours.

Slack Enterprise

"Can't find anything" complaints. 15% increase in support tickets, customer churn.

Booking.com

"Paris" query showed Paris, Texas first due to freshness signal. Social media mockery.

GitHub (Pre-2023)

Developers couldn't find code. Used external tools. Complete search rewrite in 2023.

The 5 Failure Modes

1. Zero Results

User searches for something you have, but gets "No results found."

QueryCatalog HasRoot Cause
"iPhone charger"Lightning cablesMissing synonym
"grey sweater""gray sweater"Spelling variation
"laptop under $500"Laptops at $499Price filter bug
Impact: Industry average ZRR: 5-15%. Best-in-class (Amazon): <2%. Each 1% reduction = ~1-2% revenue lift.

2. Irrelevant Results

Results exist but they're wrong.

Query: "cheap laptop"

Got: $2000 MacBook (popularity over intent)

Query: "running shoes size 10"

Got: All sizes (filter not applied)

The Position 1 Problem: 68% of clicks go to positions 1-3. Wrong #1 = failed search.

3. Slow Search

Results are relevant but take too long.

< 100msFeels instant
300-1000msAttention wanders (-2% revenue)
> 1sUser opens new tab (-5% revenue)
Amazon data: 100ms latency = 1% revenue loss ($4B/year at their scale)

4. Duplicate/Spam Results

Same product appears multiple times, or junk appears in results.

Marketplace Problem:

Query "iPhone 15 case" → 47 identical listings from different sellers. User has to manually compare.

5. Broken Filters

User applies a filter, but results don't respect it.

FilterExpectedGot
"In stock only"Available itemsOut of stock (indexing lag)
"4+ stars"Highly rated3 reviews @ 5 stars each

Quantifying the Damage

Lost Revenue = (Zero Result Searches × AOV × CVR) + (Slow Searches × Bounce Rate × AOV)

100K

Daily searches

8%

Zero result rate

$80K/day

Lost revenue

$29M/year

Annual impact

Key Takeaways

01

Zero Results = Lost Revenue

Industry average ZRR is 5-15%. Each 1% reduction drives ~1-2% revenue lift. Fix synonyms and spelling.

02

Latency Kills

>1s latency causes users to open a new tab. Amazon found 100ms latency = 1% revenue loss.

03

The Position 1 Problem

68% of clicks go to top 3 results. If result #1 is irrelevant, the user assumes the product doesn't exist.

04

Duplicates Dilute Signals

Identical listings confuse users and split ranking data, making it harder to identify the 'best' item.