U.S. retail sales are expected to rise between 3.8% and 4.4%, to more than $3.8 trillion in 2019 according to a February 2019 National Retail Federation report. These are big stakes numbers. With fast-moving market trends dictating consumption patterns, the need to anticipate customer preferences becomes the driving force for ­retailers.

To be sure, not all retailers get this right. Understanding the many drivers of a consumer-centric business involves working with a lot of data, implementing a variety of analytics and determining which insights are important enough to prompt operationalization is a big part of the ask. Fortunately, retailers have become significantly better at “doing” data science and implementing a strictly data-driven approach to making decisions about their business. The results have been very good: improved customer satisfaction, frictionless responsiveness to consumer tastes, improved efficiency in the supply chains, better price discrimination, and so on. The data science practice is now a well-funded and institutionalized entity in all the global retailers.

Retail use cases that require data science implementations

There are many important use cases in retail that require data science methodologies. While the list can be exhaustive, the top retail use cases are:

• Recommendation Engines — Recommendation engines provide suggestions for further activity based on a certain map of behaviors already undertaken. For example, people who have purchased music from a store may be given a recommendation to take music lessons at a discount, or patients who have suffered from a specific illness may be tested for a related affliction. The idea of a recommendation engine is that purchases, actions or events are related to each other based on past behavioral data. In the context of retail, since each purchase is typically correlated with others in a given set, it is easy to make recommendations when one such event is seen.

• Market Basket Analysis: A relative of recommendation engines, market baskets rely on the same concept of association analysis and a group of products bundled together for ease of purchase. The difference is that instead of individual recommendations, a group of items are pre-built into a package delivering a convenient shopping experience to customers. Market baskets provide the benefit of a one-stop shopping environment for consumers while increasing profitability for retailers.

• Inventory Management — Variables like holidays, weather patterns and supplier changes make stock keeping a challenging issue for retailers. So, the question becomes one of biting the bullet and stocking up on a lot of items or being left at the mercy of luck to ensure that demand is adequately met over time. Neither is close to optimal, because the former costs a lot in terms of inventory investment, storage costs and more, while the latter could result in incalculable reputational damage and lost customer opportunities. It is in this context that analytics is used to more accurately forecast changes in demand specifically as they are mediated holidays, weather changes and other external events. For example, with sound predictive analytics deployed, a retailer can ensure that expensive products like all-weather tires are timed to be in stores during the harsh winter months rather than clogging up precious shelf space during other seasons.

• Store Location Analysis — How often have customers refrained from purchasing products or put off purchases for a different time simply because they could not get to their favorite stores easily? Location analysis is important because it takes into consideration socioeconomic and demographic variables to determine ideal store locations. It is not a question of putting stores in places where the retailer does not have a prior physical footprint, but rather a question of understanding the potential demand and the market segments that the retailer could serve with a range of products that become important. Oftentimes, retailers do not wish to add a standard store as a part of their franchise to certain locations and may, based on the ZIP code-level analysis, choose to add specialty stores to cater to a specific demographic and/or purchasing power capability.

• Consumer Sentiment Analysis — This is one of the biggest areas of current activities among retailers: to understand prevailing sentiment and mine feedback data for new product or service ideas. Sentiment analysis typically involves natural language processing to tease out positive, negative or neutral expressions about the retailer. When mass sentiments are accurately determined, retailers can better personalize the product and service portfolio for each of their consumer segments. Furthermore, sentiments provide insights that help augment service provisions with a view towards garnering greater customer ­appreciation.

• Fraud Detection — A recent Transunion study indicated that retail fraud could reach $25 billion by 2020, with about 60% of retailers having experienced some online fraudulent activity in the last year. At least four important data points feed into fraud detection, including the following:

• Digital footprint refers to the type of device, carrier type or other media portals through which an individual undertakes ­transactions.

• Personal identity is the data provided by customers to validate their access to the retailer.

• Reputation data refers to any indicators that the retailer may use to add more scrutiny to a specific customer’s transactions.

• Transactional data is the unique pattern of purchases or interactions that indicate the likelihood of fraud. Retailers typically use all of these or a subset in consort to allow transactional participation or to altogether ban specific customers from engaging in transactions.

Challenges for data science

While it is great to wax eloquent about use cases, implementing them is far from trivial due to a wide range of issues, including:

• Not enough technically qualified people who understand what analytics solutions to implement and how.

• No bespoke solution that is readily available off the shelves (despite hearsay).

• No easy user experience that ensures different analytic personas can collaborate with each other as the analytic workflow unfolds (data ingestion, prep, analysis, model building, visualization, ­operationalization).

• Difficult to shoehorn multiple solutions (e.g., statistics, text mining, visualization) into a coherent workflow.

• No prebuilt analytic algorithms that make it easy for quick analytic execution.

Road map for action

As a direct result of these challenges, it is important to have an analytics platform that subsumes the needs of multiple analytics persona into one framework. Gone are the days when retailers had to depend on highly trained and scarcely available data scientists to do all heavy lifting and share results at the end. In today’s operationally reflexive environment, everyone in the organization needs to be data driven. This implies that analytics initiatives are undertaken with solutions that accommodate multiple technical skills (e.g., SQL users versus Python aficionados), native visualization options, the ability to ingest multiple data types at scale, operationalizing predictive models and the ability to access analytics across multiple form factors — the cloud, within a firewall and hybrid methods.

The idea is to provide retailers not with a technology challenge but to enable them to use technology with ease and to deliver impactful business outcomes. This is what will continue to separate exceptionally successful retailers from the merely good.

Sri Raghavan is in charge of Data Science and Advanced Analytics Product Marketing at Teradata.