Case Study / 2021
Product Search Platform Rebuild
Elasticsearch and multi-layer caching rebuild, cutting search latency from 480ms to 65ms and lifting conversion by 23%.
480ms to 65ms
Search latency p99
After cache and ES tuning
+23%
Search conversion
Items added-to-cart after search
78%
Cache hit rate
L1 and L2 combined
<30 sec
Index freshness
Product update to searchable
problem
Search Was Becoming a Revenue Bottleneck
The legacy search stack degraded as catalog size grew. Popular queries were slow, typo handling was inconsistent, and ranking experiments required risky deploys.
caching
Multi-layer Cache Strategy
The platform used a small in-process cache for hot normalized queries and Redis for popular result sets, with explicit invalidation when product updates changed searchable fields.
results
Measured Outcomes
P99 latency dropped from 480ms to 65ms. Search conversion increased by 23% after ranking and typo tolerance improvements shipped.
