These are unprecedented times and we are all doing our best to navigate through them together. Now more than ever, our industry is proving just how essential we are to keep things moving in reverse so that everything else can continue moving forward properly. While there are some reassuring constants, so much has changed, and sadly, some things will simply never be the same. On multiple levels, our industry is continuing to learn and adapt to this unpredictable environment. A big part of that learning process involves studying the data we have, making sense of the trends, and using that information to make informed and improved decisions.

Each movement of a product through every purchase and every return leaves a trail of valuable data that tells a story of where that product has been, what has been done to it along the way, where it ended up, and what its impacts were to the bottom line and to the environment. Helping shape that product’s journey and identity are its individual attributes such as manufacturer, originating location, descriptions, pricing, and photos. When each is assigned a meaning and analyzed in aggregate, this data becomes fundamental to making better decisions about handling the same or similar products in the future.

While capturing specific data requires careful planning and consideration, it’s good news that data collection and retention is incidental to regular process in almost all accounting, order management, ERP, and warehouse management systems in use. Whether you realize it or not, you probably already have a lot of useful data siloed away. Having that data provides the opportunity to dynamically steer dispositioning and be smarter with other decisions; a perfect application for Artificial Intelligence (AI). AI development has advanced rapidly and with the many frameworks now readily available (TensorFlow, Caffe, Thano, Scikit, Keras, etc…), putting the data to use has become a realistic endeavor and one that’s more attractive with a “critical mass” of data. AI systems are hungry and produce better results when there is more historic data to digest.

AI has a hearty appetite for data and its true genius is its ability to ingest massive amounts of it to hear the “signal in the noise. If an algorithm is a list of instructions that produce a result from the data it’s fed, AI can be thought of as a multidimensional network of algorithms looking at data and making decisions in parallel, then making adjustments where needed to improve its results the next time around. In our world, that can be applied to making profitable decisions about anything from deciding how to disposition a product at the returns counter or if and where to resell it and at what price.

The growth in the number of products flowing through the reverse supply chain has provided tremendous opportunity to collect and retain useful information. Those in the industry who have made the investment to thoughtfully curate their own data are making major strides in leveraging AI for decision-making logic. Depending on the specific purpose of the AI tool, prerequisite data to drive the desired results are different. Some of the relevant categories in reverse logistics include historic sales data and detailed product data including pricing. When the data types are weighted correctly and the right tools are used, AI can make favorable predictions about which products are worth handling, and how likely they are to be resold and where, and even the levels of velocity and recovery to expect.

Whether this process is put to use sooner or later, the investment made to capture data at every touchpoint through is well worth the resources spent to collect it. Warehouse management and ERP systems should be logging data about each product as it moves throughout its lifecycle, including key metrics like timing, who it was handled by, results from testing, transportation and freight costs, and its time spent on a shelf depreciating. Order management systems can be set up to capture data relating to sales velocities, bids and price points, previously answered questions, and location and category-based demand. Data from vendor merchant agreements, or contracts, provide beneficial visibility into potential return volumes and return allowances and caps. Finally, product level data, usually the most challenging to collect, enables decisions to be made about when and where to sell products and the information needed to identify, categorize, and remarket each one.

There are key metrics in the data that will help with decision making both upstream and downstream in the operation. Anomalies aside, AI can provide insight into estimating return volumes and planning for up/down ticks in volume with seasonality or variable/geographic demand. Using historic data, AI can also illustrate how to best disposition (liquidate/RTV/dispose) and consolidate products to achieve the highest recovery and prevent them from ending up in landfills. Together with detailed product and historic sales data, market demand, and expected depreciation, AI can predict where, when, and how to remarket items to achieve the optimal balance between sales velocity and recovery. Finally, the data from merchant/vendor agreements can provide insight into products that carry high return rates to enable better purchasing decisions in the future and also promote higher recovery by leaving dispositioning up to the retailer instead of the vendor.

As with self-driving cars, implementers of AI must be careful to know when things become too unpredictable and autopilot must be switched off. When road lanes disappear, sensors malfunction, or when the data becomes unreliable, abnormalities result in the output which can draw the wrong picture and lead to costly mistakes. Anomalies, like the one we’re all living through now, need to be carefully considered not only from the data collection side but the perspective of how it’s used by AI decision-making. AI tools are trained to operate within specific bounds by looking at the data they’re given, the results they produce and the outcome. In most cases, AI is not “aware” of surrounding factors so anomalies like this could lead to unwanted results if left entirely unsupervised. Even with no anomalies, it’s always important that the data used to make decisions is sufficient (critical mass) more than it is precise in every case.

The business of reverse logistics is complex and unpredictable by nature. When and where will the next return come from? What product will it be? What should be done with it, where, and by whom? Our ability to be savvy and successful in providing answers to these questions is directly correlated to the health of the retail sector. Thus, collecting data and being prepared to use it quickly and cleverly are crucial. Even if AI seems like a far-off project today, there’s always a benefit to making the effort to capture and preserve your data. Whether from sales history, accounting, or manifests, amassing your data and understanding the story it tells will help to drive AI powered solutions when you’re ready.

---

Larry Velman
Larry Velman has over 14 years of global information technology, software development, and technical leadership experience. As CIO at goTRG, Larry directs the development and implementation of all internal operational business and support systems as well as innovative new SaaS applications. Larry was a key influencer in the initial product design, development, and delivery of goTRG. The success of this revolutionary returns management platform defined goTRG as a leader in the reverse logistics space, and achieved record-breaking sales on third party marketplaces such as eBay and Amazon. Since that time, Larry has evolved the technical platform, executed multiple projects, and implemented technology standards, best practices, dashboarding, and reporting tools. He collaborated with senior management and cross-functional teams to establish strategy, define service levels, drive process improvement, and articulate a future vision for all products and the department. Larry graduated at the top of his class with honors in computer programming and analysis as Seneca College in Ontario, Canada. He is bilingual and speaks fluent English and Russian.