Are Your Promotions Profitable? Neal’s Data-driven Approach to Raising the Curtain

Posted By:

The traditional retail industry has been facing increasingly fierce competition and disruption by e-commence and nimble start-ups. To survive in the intensely competitive market, many retailers try to stimulate demand and attract customers by discounting products. As even high-tech gadgets become more commoditized, differentiation has given way to incentives. In one of the most intense battlegrounds, grocery promotions have doubled over the last decade. Thus, understanding promotion impact is crucial for retailers and consumer goods companies to plan effective campaigns in the future. Companies that can leverage advanced data analytics technologies to guide promotion planning are primed run more efficient and impactful promotions for each product and market segment.

The key challenge to evaluate past promotions is to understand the concept of a sales baseline. If a promotion is going to increase sales, the baseline is what would have sold if the promotion wasn’t run. Only in this way can the business evaluate the incremental impact of promotions. However, when looking at historical sales data, promotion sales periods simply report the total sales numbers, with no indication how many would have sold otherwise. To solve this problem, we isolate the past “off promo” sales data and build machine learning models to fill in the gaps of sales data when promotions were run, creating a consistent baseline for future calculations.

With the baseline derived, the next phase is to calculate various KPIs with the aim to evaluate whether a promotion is successful from different perspectives. For example, a promotion might have generated considerable amount of incremental sales volume but the incremental profit might not be able to justify the price cut and the promotion spending. Most importantly, we are after ROI, which we will use to teach a machine learning model to recommend promotions with maximum ROI for each customer/market.

The result is a series of dashboards and interfaces that allow key account managers, promotion planners, and analytics managers the ability to understand which historical promotions worked well, which didn’t, and what the best promotions to run in the future will be. Promotion calendars can be created which obviously seek to maximize profit lift while minimizing forward buying, but achieve it by planning at a much more granular level. Instead of planning “one size fits all” promotions for all stores, promotions can be planned for the demographics that like them most, resulting in happier customers and executives.

This solution is brought to life by Microsoft’s cutting-edge Azure cloud platform and data tools. We use it to achieve the agility needed to develop operationally useable solutions in this fast-changing landscape. Our Promo Optimization solution is built entirely using this platform and acts as a value driving factor in retail digital transformation. It allows businesses to get timely insights into their promotion performance and to inform further promotion planning decisions on-the-go. This powerful solution is composed of:

  • Tuned machine learning models to estimate baseline sales, evaluate impact of promotions, and forecast future promotion ROIs
  • Interactive Business Intelligence reports with analyses across regions, stores, products, and promotion mechanics
  • An automated data pipeline that allows scheduled data updates, providing users with fresh insights and continual machine learning model improvement

To conclude, our Promo Optimization solution is designed to reveal long hidden stories behind sales data and uncover the factors that make a promotion event successful. No longer do our customers need to place monumental budgets and responsibility on planners who have naught but past experience to guide them. These experts are enhanced by machine learning, using predictions of future promotion performance in their promotion calendars to drive clear and consistent improvements to promotion returns.