← BACK TO PROJECTSTwo models look at the same 10M transactions from different angles; a hybrid blends them.
Onpoint Insights
Product Recommendations Across 10M+ Transactions
A B2B warehouse distributor needed contextual cross-sell across millions of historical transactions.
HOW THE SYSTEM WORKS
DATA
10M+ transactions
extracted via SQL
STEP
Two parallel models
different signals, same data
Market Basket: “bought together”
Word2Vec: product similarity
STEP
Hybrid ranker
blends both signals per customer
OUT
Cross-sell recs
+6.7% average order size
Built
Hybrid recommender combining Market Basket Analysis, Word2Vec, and learned hybrid models.
Approach
SQL-extracted 10M+ transactions → association rules + contextual recs deployed in production.
Impact
6.7% lift in Average Order Size
Skills
Word2VecMarket BasketSQLRecommendation