To effectively transform a retail enterprise from data rich to data-led, retailers must make their data actionable. They must turn a data swamp into a smoothly running data stream. This requires clear goal setting, data normalization and seamless bidirectional communications between core systems across the enterprise. Once these basic requirements are met, retailers can effectively progress from the more basic areas of data analytics, to the most advanced:
- Descriptive: What happened?
- Predictive: What will happen?
- Prescriptive: What shall we do about it?
- Cognitive: What's the next step?
Each of these questions is important, but to different ends. As retailers embrace and evolve their approach to analytics, feeding good data into smart models, embedded retail science is delivering progressively better recommendations and forecasts.
For example, the Oracle Retail Science team, in partnership with the Massachusetts Institute of Technology (MIT), recently engaged with two different retailers to explore the application of machine learning to optimize pricing scenarios for different end goals. In the first case, a flash-sale retailer was looking to better manage inventory and maximize sell-through. The second was a large-scale retailer looking to develop an approach that would enable it to simultaneously increase revenue, profit and market share. In conjunction with MIT, we applied various models that assessed pricing, promotions and forecasting for both retailers, and ran iterative scenarios until we landed on optimal results.
In the first project, the flash-sale retailer needed to be able to maximize its first exposure selling opportunity for new inventory. To avoid having capital tied up in unsellable inventory, the desired approach needed to maximize first-exposure selling opportunity with a key metric of percentage of inventory sold, all without decreasing demand — i.e., it didn’t want to drop the price too far rendering the items undesirable or less profitable. We ran a six-month scenario and by the end of the trial, the products with pricing set by machine learning indicated an increase in revenue by 10 percent to 12 percent (with a 90 percent confidence interval), while the most expensive items saw an increase of 22 percent — all while maintaining market share.
In the second project the large-scale retailer was looking to increase revenue, profit and market share for premium products. With multiple KPIs that can negatively impact one another, this project had three main challenges: demand curve generation, product clustering, and efficiently solving for optimization. Our team tested multiple approaches, ultimately landing on a three-pronged approach to solve for each challenge. Results for premium product sales governed by the algorithm were spectacular: revenue up 471 percent, profitability up 366 percent and unit sales up 391 percent. From a competitive standpoint, dynamic pricing is a powerful tool for market positioning. In this case, the retailer was able to sell 40 percent more unique SKUs every day vs. the control group of SKUs.
Although both cases were for online retail, similar tactics can be applied to any channel — brick-and-mortar, social, catalog, etc. Applying advanced retail science, regardless of sales channel, can have dramatic improvements in core KPIs, including profitability, sell-through, revenue and market share. For all retailers looking to evolve their operations from data-rich to data-led and up level their analytics game from level one or two to level three or four, we strongly recommend collaborating with a partner that has the breadth and depth of past learnings to increase revenue without running the risk of adversely affecting market share.
The full the results of the study with Oracle Retail Science and MIT can be found within the guidebook, Innovation in Retail: Using Machine Learning to Optimize Performance.
Troy Parent is senior director, retail science and insights, Oracle Retail.