As part of the 2015 Optimization Summit held in Bangalore, I had the chance to present the science contributions underlying the JDA Retail.Me initiative introduced to the public in January at NRF.
Why does the JDA Labs’ Science team love retail so much? It’s not just because we like shopping, but rather because there are numerous opportunities for science. Retail is shifting from separate channels to omni-channel, from being product centric to customer centric, from static assortments to fast changing assortments and collections, and new challenges and complexities keep arising. As those challenges come with a fast, continuous flow of data from various sources, they bring a critical need for automation and analytics. The science behind Retail.Me, along with advanced user experience and development technologies, has one goal– support user decisions by letting them focus on the ‘art’ while the system takes care of the data crunching to support it.
How can science help the user with decisions such as who shops in my different touch points and what items do they like? What, when and where should I carry which items?
First, we segment customers in order to reveal actionable insights. The key is to find out which customers have the most value for the organization based on, for example, their total spend, basket size, proportion of full price purchase, their fulfillment behavior (do they buy online for home delivery or pickup in store), percentage of returns, loyalty, social influence, etc. Then, we highlight where they are and what their preferences are to get the assortment right for them.
Building the right assortment for your most valuable segments means knowing how much will an item appeal to them, even if that item and most of the other items that are considered for the assortment have never been seen before. We use advanced machine learning approaches to measure how an item will appeal to different customer segments, to suggest what else should be in the assortment at the same time, and what the lifecycle plan should be for those items. Even if the considered items and assortments are new, we use patterns from the past to inform us of the expected future behaviors of customers towards those items.
Another major challenge occurs when the assortment hits different touch points. Since uncertainty is present in any operational stage, aiming to achieve the perfect plan by assuming perfect information is impossible. To break the silos between planning and execution, a system that can interpret uncertain information must be put in place. Such a system contains the following two components and eases the execution during trading or in-season:
- Learn patterns that announce a deviation from the plan to alert the user on a real-time basis to have choices of actions in place;
- Have actionable scenarios with predicted outcomes ready for the user to decide quickly, based on what will be observed.
The underlying algorithms are similar to the ones supporting planning decisions, but with refined outputs learning from what is actually happening.
The fundamental problems described above represent only the tip of the iceberg of what science can bring to retail. As more challenges emerge, decision and data analytics will become more central to day-to-day retail operations. To be ready, the first step for retailers is to collect and store all the data surrounding their business. As I said earlier, we data scientists love retail, but we love data even more!