ML-driven price and promotion optimization on Azure: 8 weeks Implementation

Grid Dynamics Holdings, Inc.

Create an ML platform on Azure that leverages sales data, product attributes, competitor pricing, and external factors to intelligently optimize prices and promotions.

We help companies to create an Azure ML pipeline that leverages historical sales data, product attributes, competitor pricing, and information about various external factors to automatically optimize list prices and promotions.

This pipeline can be integrated with the existing price management and ERP systems to improve the quality of pricing decisions and automate the workflows.

The solution is developed based on a reference implementation (starter kit) which is customized to meet the need of a specific customer. The primary goal of the Price Optimization Starter Kit is to reduce the costs, timelines, and risks associated with the development of price management solutions without compromising on the flexibility of the open source based approach.

  1. Designed by revenue management experts and data scientists who specialize in price optimization solutions according to the industry best practices.
  2. Implements the reference price optimization pipeline using only open source Python libraries and cloud native services.
  3. Can be deployed and integrated with the real data sources in a matter of hours, and then evaluated and customized based on the unique requirements of the company. No proprietary components, no licensing costs.

The Price Optimization Starter Kit provides a reference pipeline that includes the following steps:

  1. Data preparation. Sales and catalog data are loaded and transformed in a format suitable for the downstream modeling.
  2. Demand forecasting. The demand forecasting model is used to evaluate the demand for different pricing scenarios.
  3. Demand analysis. The dependencies captured by the demand forecasting model are visualized to provide insights into the demand structure.
  4. Price and promotion optimization. The forecasting model is evaluated for a grid of pricing parameters and profit-optimal prices are determined for each combination of a SKU and date.


  • The solution is designed to support thousands of SKU.
  • Demand forecasting model account for SKU attributes which enables transfer learning across the products.
  • The solution supports modeling of cross-product effects such as cannibalization and halo.
  • The solution supports custom profit calculations and optimization objectives.