B2B ecommerce supply chain process improvement leveraging cross-channel data


Client is a B2B software platform that helps brands selling on Amazon grow their ecommerce sales by tracking, monitoring, and optimizing various key marketing metrics (related to ecommerce sales, campaign effectiveness, cross-channel pricing, consumer behavior etc.). They were using Amazon’s forecast on Vendor Central portal for each vendor for each ASIN in order to provide sell-through forecasts and sell-in recommendations for better inventory planning and purchase order fulfilment. However, it was observed that the accuracies of those forecasts were not satisfactory. Consequently, the client wanted to leverage cross-channel data (e.g., campaigns response data, web behavior data, online reviews statistics, promotions data etc.) already being curated by them, in addition to past sales data, and utilize machine learning for more accurate and refined forecasts and recommendations for their customers.

B2B ecommerce supply chain process improvement leveraging cross-channel data


Business Understanding and Data Enrichment Strategy

We collaborated with the client to draw up a holistic list of parameters which might impact the demand and worked with them to establish the corresponding data enrichment strategy involving those attributes in various curated data tables. The ABC model of prioritization was used to finalize the target vendors and ASINs which were found to be more important based on business KPIs.

Data Enrichment

An enriched data table was created that was amenable for feeding into machine learning models. The enriched table involved factors related to:

  • Web Behavior (attributes such as Glance Views, Conversion, Lost Buy Box % etc.).
  • Online Reviews (attributes such as Relative Negative Reviews).
  • Promotion (attributes such as AMS Spent, Prime Day and other Special Days).
  • Pricing (attributes such as Average Buy Box Price, Price Discount, Best Deal etc.).
  • Historical Sales and OOS (Out of Stock) data.

Model Development

Business-aware feature engineering followed by custom ML model was developed to forecast the sell-through demands for target vendor-ASIN combinations. This forecast finally was combined with Inventory and OOS information in the recent past to produce sell-in recommendations over the next 26 weeks lookahead.


Sell-through forecast accuracies improved by ~15% across the target vendor-ASIN combinations. The sell-in recommendations were adopted by the major target vendors.

Case Studies