Thursday, January 2, 2020

Predict the Impact of Retail Sales on Revenue Using Azure Databricks



Business Case:

Our client, a team of sales managers at a multinational technology company, sells software products through retailers. Our client needed to forecast the impact of retail sales on their overall revenue. With increased insight, our client could proactively focus on retailers with higher sales impact.

Key Challenges:

  Forecast revenue using existing data sources 
  Decrease forecast run time 
  Enable scalability for future business growth 

Our Solution:

We created a machine learning (ML) forecast model using Azure Databricks.

Figure 1: Solution Design

We collected data from existing customer relationship management (CRM) and sales systems. Including data sets, such as customer geography and the status of deals between clients and dealers, would increase forecast accuracy.

 

The collected data was input into Azure Databricks, in which we created a forecasting model using existing and custom linear regression. Originally, the model was based on existing data libraries of “R” environment, which slowed forecast run time. We modified the “R” libraries to “Spark R”, enabling the forecasts to run faster. Using a cloud-based platform like Azure Databricks enabled scalability and near real-time analytics.

Business Outcomes:

    Increased cost-efficiency and business growth potential with scalable Power BI forecasting.

    Reduced forecast run time from 3 hours to 15 minutes, achieving near real-time analysis.

    Enhanced security through role-based access control, so that current and forecasted revenue data is safe from unauthorized access.

•  Successfully integrated existing data sources and increased forecast accuracy with added data points