Through innovative products and experiences, retailers have historically harnessed change to their advantage. However, there has been a paradigm shift, and retailers are now undergoing a phase of disruption of their own that is forcing them to change how they do business.
Many forces have contributed to this shift, including technology, demographics, and changing consumer purchasing habits. Retailers are being battered, which is leading to declining profitability. Fortunately, retailers have tools at their disposal that can help them turn this declining trend around. Predictive analytics leverages machine learning to generate incremental revenue quickly.
The current state of retail is such that retailers with store networks rely on any number of back-end platforms that are excellent for financial planning and aggregation of historical data for various uses. In other words, these systems provide great business intelligence that looks into the past. In general, however, these systems cannot produce robust future outcomes. As is the case in finance, past results are not always indicative of future results.
There is an abundance of data in retail that comes in many forms, but traditional analytical methods do not make the most productive use of data in a prescriptive manner that retailers can execute against. There is a lot of complexity in retail. Therefore, in the current state of affairs, human intuition plays a big role in the decisions retailers make. However, combining human intuition with predictive analytics that leverages machine learning can yield better outcomes. In essence, predictive analytics can help guide human intuition to help improve financial results.
There are many mathematical techniques that can be used in predictive analytics to find answers using retail data. One of the most common is the classic statistical method based on cause and effect, linear regression. For example, linear regression can help answer the question: If the advertising budget is increased by 5 percent, how much will sales increase next season? This logic assumes that advertising drives sales. More variables can be added to see what else, besides advertising, might drive sales, but the outcome will always be based on a linear model. However, there is another mathematical technique that is better suited for the nonlinear complexities of retail— machine learning.
With machine learning a future question can be answered by taking into account millions of complex possibilities in a nonlinear manner. Machine learning uses algorithms—very simply, a set of rules—to process data, learn from that data, and make future predictions. This allows for a more robust predictive model, which can continually improve over time. In this context, questions like, “What should a future fall/winter buy look like?” can be answered. When coupled with machine learning, predictive analytics is immensely valuable in retail. Nonlinear insights that may not be obvious will be uncovered.
Adapting to Customer Demands
The retail environment is changing rapidly and dramatically. Consumers are demanding personalization, convenience, and speed at all touch points, including the digital and brick-and-mortar channels. Digital devices and demographics (i.e., millennial consumers) have changed how customers interact with retailers. In addition, the growing reliance on and importance of mobile devices in their lives has also shaped ways in which consumers expect to interact with retailers.
At a high level, all of this seems obvious. However, omnichannel interactions are extremely tough for most retailers to achieve. The current retail model was originally conceived for physical brick-and-mortar, not digital commerce, which has arrived much more recently. Newer retail companies are not encumbered by the older model and are often more nimble because they started from scratch and created fresh retail models. Although digital continues to increase as a percentage of overall retail revenues, retailers still need to “mind the store” and pay very close attention to their brick-and-mortar outlets. According to studies, between 80 and 90 percent of revenues are still generated by brick-and-mortar stores. From a consumer standpoint, there is no difference between the physical and digital segments of a retail business.
Predictive analytics can help retailers remain competitive. Amazon has a long history of innovation in the retailing domain—e.g., Amazon Prime, customer reviews, 1-click ordering, a sophisticated fulfillment network, etc. The company has been using and innovating with machine learning for years. Although Amazon is still mostly an online retailer, it has been slowly moving into the physical realm as well. In December 2016 the company unveiled Amazon Go, an “experimental” self-checkout convenience store. In March 2017 it opened its 10th bookstore and is expected to open even more stores going forward. All retailers should take notice that Amazon is now entering the traditional brick-and-mortar domain of retail.
Predictive analytics lends itself for use in a variety of retail scenarios in brick-and-mortar retail. Predictive analytics can be used to address specific problems in assortment optimization, pricing, markdowns, marketing, and online order management from retail locations and many other areas. This article concentrates mostly on assortment optimization. Why? Localized optimized retail product assortments result in increased revenue, higher inventory turns, reduced markdowns, and significant increases in customer satisfaction. Astute executives ensure that they are being highly productive with their inventory, while controlling expenses.
Optimizing In-Store Product Assortments
The golden retail rule is: get the right product in the right place at the right time at the right price. Merchants and planners have always strived to reach this goal when making purchases. Part of what makes this extremely difficult is the fact that retailers often have sparse data on individual customers, assuming that a customer is being identified when making a purchase via any sales channel, whether physical or digital, in the first place. In other words, if a customer visits a given retailer two or three times a year, there is very little data from which to make precise granular inferences for the future with the tools that are in use today.
Hence, the data that is aggregated and being used to drive decision making for future purchases is not processed to provide true insights. It is a view of the past and is generally fit into a model that groups stores by revenue/size, e.g., A, B, C; doors; and region, e.g., Northeast, Southwest, Middle States, etc.; or some variation of this. The reality is that today, most retailers are making multimillion-dollar, forward-looking purchasing decisions based on Excel spreadsheets and gut instinct.
When applied to retail, advances and innovations in machine learning can help a retailer increase revenue incrementally when purchasing and planning assortments for future seasons. For example, a women’s dress department that is part of a retail store chain that has 60 stores, 10 deliveries a year, and a budget of $500 million might purchase 70,000 dress SKUs (style, color, size) in a year. There is an additional constraint that, on average, each store has room for 3,000 dress SKUs.
How would a team begin to whittle away at this problem and come up with a viable strategy that maximizes revenue? Until now, merchants and planners did not have the tools that allowed them to do their jobs with great precision. Now blind spots can be easily considered and specific business questions like these can be quickly answered: Is the retailer investing in the right styles? Is it overbuying or underbuying? Did it allocate the right product in the right stores?
As much as technology has changed how consumers shop, it has also benefited retailers by providing a better way to model future outcomes that are not constrained by old mental models. When these data are aggregated and subjected to machine learning and analytics computations, the resulting customer choice model can accurately predict future customer intent.
With current tools, any modeling has to be done manually with Excel spreadsheets. For example, creating an optimal store assortment for a single store may take days using spreadsheets, whereas creating an optimal store assortment with machine learning takes only a few seconds. Moreover, since each department—e.g., men’s, women’s, accessories, home, jewelry, etc.—perform planning calculations separately, crucial subtleties are lost in the process. For example, machine learning can automatically factor in the effect of complementary or substitute products and come up with optimal future outcomes. This is simply not feasible with the tools in use today.
Other retail applications include instantaneous vendor comparisons that visually demonstrate business potential from the high level—store, region, department—to the most granular level, such as fabric, color, and style. It is now even possible to create distinct and precise clusters based on shopping patterns, rather than the A,B,C revenue/size and region models.
Not a Panacea
Although predictive analytics can help in many areas of retail, there must first be a clear strategic road map from the C-suite before it can be applied effectively. Predictive analytics cannot just be thrown into the mix. It must be incorporated into a well-crafted, thoughtful strategy and used in clearly defined areas that address some of the most pressing problems, such as inventory. Any new initiative or solution must answer a specific business question.
Additionally, it is important to remember that technology, on its own, cannot solve all business problems. Moreover, to achieve long-term success, the overall strategy needs to look beyond quarterly results. Amazon has taught The Street to not expect immediate profitability. This has given the company flexibility to innovate and dominate many of the markets it enters.
Finally, it is important to remember that when applying predictive analytics to problems, there must be an element of human intuition factored in, as this creates a more balanced approach. Predictive analytics complements human intuition; it is not a replacement for it. The current retail phase of disruption will create winners and losers. Retailers should have the right predictive analytics tools to improve their chances for being on the winning side.