All case studies
Retail
AI Recommendation Engine Custom AI Solutions
A national online retailer with 50,000+ SKUs was still running a five-year-old recommender - two-thirds of its "Recommended for you" slots were out-of-stock items or repeat purchases.
Retail
Industry
9 tools
Stack
Client
RetailMax
Industry
Retail
Service
Custom AI Solutions
Stack
PyTorch, Sentence Transformers, AWS SageMaker
01
Challenge
“A national online retailer with 50,000+ SKUs was still running a five-year-old recommender - two-thirds of its "Recommended for you" slots were out-of-stock items or repeat purchases.”
02
Build
We rebuilt the recommendation layer as a hybrid model - collaborative filtering plus a session-intent transformer plus a real-time inventory and margin signal - served from a feature store under 50ms, with a daily evaluation gate before any model ships.
03
Outcome
3.2x conversion lift on recommendation surfaces and out-of-stock recommendations down from 31% to under 2%.
Deliverables
What the system does — functionality shipped.
- 3.2x conversion lift on "Recommended for you" surfaces. 42% increase in average order value among customers who interact with recommendations. Out-of-stock recommendation rate dropped from 31% to under 2%. Model retraining and shadow-evaluation pipeline owned by the in-house data team.
Technologies
PyTorchSentence TransformersAWS SageMakerFeature Store (Feast)Real-time inference (FastAPI)SnowflakeAirflowReactA/B testing
Similar Build?
Tell us your workflow. We'll review with this example as context and suggest the best path forward.
More Work
View all
Case study © AorBorC Technologies · aorborc.com · +1 (872) 267-2672 (872-AORBORC)