GCP · Retail · E-commerce
Real-time personalization engine drove +18% conversion for a global retailer.
We rebuilt the personalization engine on GCP with real-time feature streaming and Vertex AI — +18% conversion and 3× peak-season scale.
[Conversion]
+18%
[Peak scale]
3×
[Recs latency P95]
<80ms
[ The Problem ]
Where the business was stuck.
Personalization was batch-computed nightly — irrelevant by the time shoppers saw it.
[ Key Challenges ]
- ▸Nightly batch recommendations went stale by morning
- ▸Peak-season traffic broke the recs service
- ▸No unified feature store across channels
Our approach
How we engineered
the outcome.
STEP 01
Real-time feature store on BigTable + Dataflow
STEP 02
Recs models on Vertex AI with online eval
STEP 03
Serverless storefront components on Cloud Run
[ Solution highlights ]
Delivered — measured — in production.
- ✓Conversion up 18%
- ✓3× peak-season traffic served comfortably
- ✓Unified feature store across web, app and in-store
[ Tech stack ]
Built on GCP.
GCP
Vertex AI
BigTable
Dataflow
Cloud Run
"
The store finally feels personal — in real time, at peak.
Keep exploring