Ever wonder how Netflix gives you recommendations for the next movie or how your smartphone knows that you will be driving to work Monday morning? Those are both examples of machine learning.

Machine learning is a subset of artificial intelligence. It creates computer-based algorithms designed to learn from data—and adapt to new data—without being explicitly programmed. In the past, using traditional methods such as regression analysis, an analyst had to define the objective and look for correlations between that objective and a defined set of data inputs. If new data came in, the analyst needed to rerun the analysis, create new correlations and develop a new algorithm. Machine learning basically alleviates the need for manual intervention with computing programs that automatically take new data inputs and “learn” from them.

 

Many companies are already exploring this technology. In their March 2017 “Customer Experience Innovation Survey—AI now on the CX Map” report, Gartner surveyed a panel of IT and business leaders. The data indicates that, “more than half of the organizations in the 2017 study expect to be using machine learning within the next three years, 47% expect to be using virtual customer assistants or chatbots, and 40% are focused on virtual personal assistants for their customers.” According to the Gartner report, 54% of businesses surveyed stated that they have either implemented machine learning or have plans to do so in the next 1–3 years.¹

A compelling, personalized offer can go a long way in influencing a consumer’s shopping and buying habits. Keep in mind that contacting a consumer has a cost, both financially and possibly in reputation, because a poorly targeted offer can be annoying. Using machine learning, a company can decide which customer should be targeted with what offer, better addressing their needs and wants, in addition to eliminating wasteful and costly marketing.

In the retail environment, machine learning starts with SKU level data—the individual items purchased in each customer transaction. Data scientists can store this information on a massively parallel processing database, with multiple processors carrying out coordinated computations simultaneously. Using NBO models, they can more efficiently develop a list of probabilities that a consumer will shop in one of several predefined categories, depending on the retailer. For example, a model built for a mass merchandise retailer can predict whether a customer’s next purchase will likely involve automotive goods, baby care items or housewares.

Inputs are provided by information, such as what the consumer buys next and/or how they rate the offer presented. These inputs allow the model to continually modify probabilities. This continuous process integrates new information, closing the learning loop so the customer can effectively “tell” the company what he or she wants based on past behaviors and present preferences.

Machine learning’s data-driven intelligence can identify gaps in performance efficiency by understanding how a location (e.g., store front, gas station) should perform. A technique called Data Envelopment Analysis (DEA) can show a retailer why a store with the most sales isn’t necessarily the most efficient one. How? Basically, DEA is a method that can compare a store’s various inputs using hundreds of characteristics—such as the number of customers, employees and nearby competitors—to determine its efficiency. From there, the algorithm ranks each site based on the differences between the expectations and the actual results to come up with an efficiency ranking. Based on that number, a retailer can determine which stores need the most help—and the most appropriate type of help they need.

Comparing the efficiency of various locations can be challenging at face value. One location could be newer—with upgraded amenities and square footage—but located in a less trafficked location. In the past, it was more difficult to determine if this location should overperform or underperform a smaller, older location with fewer amenities but more foot traffic. By using machine learning, thousands of characteristics can be entered in numerous combinations to determine which store is performing above or below expectations.

 

Machine learning not only allows retailers to estimate what a given location should produce, but also lets retailers rank stores at a very granular level. In a market like Chicago, for instance, with numerous locations, a retailer can cluster their stores into neighborhoods and use machine learning to prioritize where they should spend their time and money.