While pricing is recognized as a pivotal determinant of success in the retail industry, it is also one of the most challenging merchandising tasks. Retailers face many challenges when choosing pricing strategies to maximize profit, including accurately forecasting financial impact of potential pricing tactics, taking reasonable consideration of core business constraints, and fairly validating the executed pricing decisions. Expanding product offerings add further computational requirements to make in-time pricing decisions, compounding the difficulty of this already overwhelming task.
The solution provided here addresses the challenges raised above by utilizing historical transaction data to train a demand forecasting model that predicts the impact of store, department, brand, and product attributes on demand and sales rates. Pricing of products in a competing group is also incorporated to predict cross product impact such as cannibalization. A price optimization algorithm can then employ the model to forecast demand at various candidate price points while considering business constraints such as feasible price ranges, and choose the combination which maximizes profit. An experiment on store level is designed to evaluate algorithm performance, compared to the alternate pricing strategy. The whole process described above is operationalized and deployed in the Cortana Intelligence Suite.
This solution will enable retailers to ingest historical transaction data , predict future demand, and obtain optimal pricing recommendations on a regular basis, consequently improving profitability and reducing the time and effort required for pricing tasks.
For a discussion of the analytical approach used in this solution, see the Solution Description in the Manual Deployment Guide.
In this session, we provide more details about how the above analytical approach is operationalized in Cortana Intelligence Suite. The following chart describes the solution architecture.
Raw simulated transactional data are pushed into Azure Data Lake Storage, whence the Spark Jobs run on HDInsight Cluster will take the raw data as inputs and:
- Turn the unstructured raw data into structured data and aggregate the individual transactions into weekly sales data.
- Train demand forecasting model on the aggregated sales data.
- Run the optimization algorithm and return the optimal prices for all products in all competing groups.
The final results are visualized in Power BI Dashboard. The whole process is scheduled weekly, with data movement and scheduling managed by Azure Data Factory.
A parallel version of the price optimization algorithm is implemented on Spark. Utilizing RDD.map(), the independent price optimization problems for products in different competing group can be solved in parallel, reducing runtime.
The snapshot below shows the Power BI dashboard that visualizes the results of retail price optimization.
The dashboard contains four parts:
- Price Elasticity: shows the relationship between sales and price, and using the filters on the right, you can select to view the results for a specific store, department or product.
- Demand Forecasting: shows the results and performance of the demand forecasting model.
- Price Optimization shows the profit gain realized by using the recommended optimal price, as well as corresponding changes in sales volume and price that resulted in the profit gain.
- Execution Time shows the time decomposition of different computational stages, allowing the user to monitor the runtime.
This solution template contains materials to help both technical and business audiences understand our price optimization solution for the retail industry built on the Cortana Intelligence Suite.
In this repository you will find a folder labeled Solution Overview for Business Audiences. This folder contains:
- Walking Deck: In-depth exploration of the solution for business audiences
For more information on how to tailor Cortana Intelligence to your needs, connect with one of our partners.
See the Manual Deployment Guide folder for a full set of instructions on how to deploy the end-to-end pipeline, including a step-by-step walkthrough and files containing all the scripts that you’ll need to deploy resources. For technical problems or questions about deploying this solution, please post in the issues tab of the repository.