01 · Four years of transactions
AWS Glue and Amazon EMR prepared about 20 GB of transactions spanning four years in Amazon S3.
Customer work · Distribution
A dealer recommendation system increased click-through rate by 15%
A distributor wanted to improve product discovery across a 16,000-dealer network. BluePi prepared catalogue, dealer, purchase, and SKU data on AWS and built recommendation paths for related products, similar-dealer behavior, and seasonal stocking.
Dealer recommendations for app and web ordering
The distributor served 16,000 dealers through app and web portals, but personalization was limited and behavioral data covered purchases without navigation events. BluePi prepared catalogue, user demographic, dealer, SKU, and purchase-history data on AWS and built recommendation paths for product discovery and stocking needs.
02 · Three discovery routes
Recommendations used related and frequently co-purchased products, similar-dealer purchases, and seasonal demand for the next two months.
03 · Ordering portal delivery
Recommendation sets reached the dealer app and web experience and increased click-through rate by 15 percent.
Recommendations built from catalogue, dealer, and purchase records
Catalogue, user demographic, dealer, SKU, and purchase-history data landed in Amazon S3. AWS Glue and Amazon EMR prepared the transaction record for modelling, and recommendation logic used HRNN, collaborative filtering, SIMS, and time-series forecasting. A separate documented variant used Amazon Personalize.
Prepare the source data
Catalogue, dealer, user, SKU, and purchase records were transformed into model-ready inputs on AWS.
Build three recommendation paths
Related products, similar-dealer behavior, and seasonal stocking addressed distinct discovery needs.
Return recommendations to ordering
Recommendation sets were delivered through the dealer app and web experience.
Where this pattern fits
This case is relevant when a distribution network has purchase history and catalogue data but limited product discovery or behavioral context.
Working-session promptsQuestions that define the delivery boundary
Use these prompts to decide whether the case fits your operating problem and what a first deployment should prove.
- Which catalogue, dealer, user, SKU, and purchase records are available?
- Which recommendation need matters first: related products, peer behavior, or seasonal stocking?
- How much history is representative for each dealer and category?
- Should the implementation use custom models or a managed recommendation service?
- Which interaction establishes value for the ordering experience?
A practical first step
A first engagement can prepare one catalogue and dealer segment, build one recommendation path, deliver it through a controlled portal experience, and compare click behavior against the existing discovery route.
Case details
Open a section to review the customer problem, implementation, business change, and architecture.
01Starting point
Dealers placed category orders through app and web portals, but the experience offered limited personalization. Behavioral data covered purchase history without navigation or exploration events, constraining how the system could understand current interest.
- 16,000-dealer network
- Approximately 20,000 users
- Approximately 20 GB of transactional data spanning four years
02What BluePi built
Product catalogue, user demographic, dealer, SKU, and purchase-history data was staged in Amazon S3 and prepared for recommendation modelling. The system produced three recommendation sets.
- Related and frequently co-purchased products
- Products selected by similar dealers
- Products expected to be relevant during the next two months
- AWS Glue and Amazon EMR data preparation
- HRNN and collaborative-filtering recommendation paths
- SIMS category recommendations
- Time-series forecasting for seasonal products
03Managed-service variant
A second implementation path prepared the same source data in Amazon Personalize format and built content, category, and seasonal recommendation use cases with Amazon Personalize.
04Recommendation outcome
The recommendation experience increased click-through rate by 15%.