The retail trade has been around for thousands of years. For most of that time, local merchants brought in products to meet the needs and wants of their town’s population.
The face of retail has changed unrecognizably since those days, for better and worse. Consumers have more choices than ever before, and a shopping experience their forebears would marvel at. But something precious has been lost in the process. With so many locations, so many options, and so many factors affecting consumer preferences and demand, today’s merchants are simply overwhelmed. Retailers all too often miss the mark, as evidenced by unacceptably high levels of out-of-stocks, markdowns, abandoned shopping carts and many other disconnects between retailers and their customers.
The traditional approach to forecasting is just that — traditional. Time-series based forecasting, the basis for nearly all of today’s forecasting applications, dates back to the early 1960s, which means most merchants are basing their forecasts on science that’s older than them. Unlike music, technology does not stand the test of time.
The antiquated nature of forecasting science used by the majority of retailers creates major disconnects between true demand and what ultimately ends up in their assortment and inventories. It requires merchants to somehow manually apply their local market knowledge to enhance the crude numbers generated, without any context, by legacy systems. It drives up costs and frustration throughout the enterprise, and leads to overall unprofitability and inefficiency across operations.
Enter machine learning-based forecasting. Just as IBM Watson — perhaps the best known example of machine learning — can replicate the perfect doctor who can accurately diagnose even the most complex symptoms within seconds, machine learning can help retailers quickly and accurately understand and forecast consumer demand, in ways that otherwise are impossible. It’s the same type of technology that helps giants such as Amazon and Netflix understand consumer preferences and make accurate suggestions.
With machine learning-based forecasting models, retailers can examine all the drivers of demand — relieving the burden on merchants who otherwise must manually adjust each forecast. Machine learning-based forecasting systems can automatically evaluate and prioritize insights based on all the attributes that affect demand, allowing for a far better informed decision-making process, and ultimately more profitable outcomes.
With machine learning, merchandising, marketing, operations and supply chain teams can be armed with accurate and detailed forecasts — and a previously unattainable understanding of the underlying drivers of consumer demand — that can dramatically improve the organization. And we’re not just talking about the “future of retail” either. Machine learning is already in use today. Here are just a few examples:
Fresh food losses are a multi-billion dollar a year problem across the grocery industry — yet a leading international grocery chain has applied machine learning-based forecasting to demonstrate how to improve fresh food forecasts and reduce waste by 50%.
In another example, a national top-20 retailer that applied cloud-based demand forecasting techniques to reduce inventories by $80 million a year is now improving forecasts further by applying machine learning to data culled from competitors, operating markets, regional weather, demographics and economic conditions.
What’s at stake for retailers? Even a 5% improvement in forecasts can result in a 3% reduction in inventories, with similar increases in sales and margins. It’s possible to see improvements of 25% to 50%, particularly at the local level and for the most challenging forecasts, such as promotions and new items.
As retailers look for the best resources to help them compete in an increasingly digital world that caters to consumers’ multi-faceted lifestyles, it’s critical to harness the power of advanced technologies. Machine learning is a disruptive technology that is fundamentally designed to uncover all the drivers of demand. Moreover, machine learning holds the key to recovering retail’s lost vision of the merchant that personally knows each customer and is perfectly positioned to anticipate and meet his or her needs.
Ron Menich is EVP and Chief Scientist at Predictix. He previously served as Chief Scientist of Pricing and Revenue Management at JDA, where he was the conceptual designer of many of the core modules in JDA’s Pricing and Revenue Management suite. Menich is a proven professional with 18 years of experience in demand forecasting, optimization and large-scale recommendation processing systems. He brings to Predictix unmatched experience in leading a science team and designing advanced software solutions.