What Are the Risks for Retailers When Adopting Artificial Intelligence?

What Are the Risks for Retailers When Adopting Artificial Intelligence?
November 1 , 2021  |  By Lucas Cuni-Mertz, Hyunseok Jung

Share this via:

Retail is one of the largest industries in the United States, generating $5.4 trillion worth of sales in 2019 alone. With giants like Amazon, Walmart and Target, the retail industry is an essential part of many Americans’ lives. 

Broadly speaking, these retailers can boost profits in three ways: by increasing in-store sales, increasing online sales, and/or improving supply chain efficiency.  For this reason, many retailers are excited about the growth of artificial intelligence (AI), technology that can accomplish all of these goals and more. For example, artificial intelligence can improve mobile shopping by personalizing its recommendations to each customer, help maintain a better in-store experience, and improve payments, customer service, logistics and inventory optimization

We can already see numerous examples of AI’s impact on retail. Using smart-shelf technology, Kroger delivers customized product offers and pricing to shoppers as they walk through their local grocery store. At restaurants like McDonald’s and Taco Bell, customers can now order on touchscreens rather than at the register. And at Amazon Go stores, an automatic check-out process allows customers to purchase items by simply picking them up.

While many retailers have adopted AI, these applications can pose risks to retailers if implemented poorly. In their article “How artificial intelligence will affect the future of retailing,” Abhijit Guha, Dhruv Grewal, Praveen K. Kopalle, Michael Haenlein, Matthew J. Schneider, Hyunseok Jung, Rida Moustafa, Dinesh R. Hedge and Gary Hawkins examine the risks associated with adopting AI, including factors such as whether the application is customer-facing, the amount of value it can create, whether it’s online or in-person, and possible ethical concerns

Customer-facing vs Non-customer-Facing Artificial Intelligence

The primary risk factor retailers should consider when adopting AI, the authors argue, is whether the application is “customer-facing.” As the name suggests, customer-facing applications directly interact with customers. These include personalized recommendations and in-store customer management (e.g., Amazon Go). By contrast, non-customer-facing applications primarily deal with tasks the customers don’t see, such as inventory optimization and logistics (e.g., Giant Eagle store’s Tally robot). 

When implementing AI, retailers balance risk (e.g., possible customer backlash) and return (e.g., commercial benefits). Because customer-facing AI applications are seen by customers, the risks associated with their implementation are greater, the authors argue. Retailers are thus more likely to adopt non-customer-facing AI applications than customer-facing ones

Additional risk factors also influence a retailer’s decision to implement customer-facing or non-facing AI, which are explored below. 

Application Value 

 Out of 19 industries,  AI  is predicted to have the most value impact on the retailing sector. This value potential is due to various factors. First, retailers frequently interact with customers, providing stores with customer data such as purchase history and demographic information. Second, this data can be augmented with second-party data from “retailer-linked sources” (e.g., social media data). This vast amount of data retailers collect is perfect for AI to capitalize on. According to the authors, “When AI is used to analyze the sum totality of such data, some AI applications may have the ability to deliver high-value predictions and recommendations.” 

Whether an AI application is adopted is moderated by this value potential, the authors argue. Retailers are more like to adopt non-customer facing AI applications (as compared to customer-facing applications), but less so if the customer-facing application has substantial value potential.

Online Versus In-store Use 

Many AI applications are used both online and in-store. However, the way we view these applications can be quite different in these settings. For example, when shopping on the Walmart app, you likely wouldn’t find it strange if it recommended products based on your purchase history. However, you might feel uncomfortable if you received similar recommendations via text message from Walmart while shopping at one of their stores or simply driving by one. “Inside a retail store,” the authors note, “there is more significant potential for customer discomfort, as customers often have a greater expectation of privacy than they do online.” 

Because of this, the authors predict that retailers are likelier to adopt “online setting” AI applications, as customers often have fewer privacy concerns online and view the AI as less intrusive. Conversely, retailers are less likely to adopt in-person AI applications due to the greater discomfort associated with them.  

Ethical Impacts of Artificial Intelligence  

The ethical impacts of AI applications are also important for retailers to consider. Specifically, concerns about data privacy, bias and the application’s appropriateness can affect if and how AI is implemented. 

Concerns about privacy. AI applications are increasingly able to identify specific people based on limited data. Of course, a customer is identifiable if they provide their full name or credit card information to a company, but they can also be personally identified by AI even if they only provide broader information like age, zip code, or gender. 

The sensitivity of consumer data matters as well. For example, a company having data on your grocery shopping habits  might not concern you, but this may not be the case if they knew about sensitive purchases like birth control. The authors observe that “The identifiability and sensitivity of customer data are likely to be higher with AI applications…and so customers’ privacy concerns may well be heightened.” These concerns, they say, are primarily associated with customer-facing applications, meaning that privacy concerns may prevent retailers from adopting such applications

Concerns about bias. AI’s ability to enforce biases from the data it learns is another concern. “For example,” they write, “even if race or gender is not a formal input into an AI algorithm, an AI application may impute race/gender from other data and use this to ‘price higher’ to specific demographics.” Apple learned this the hard way when its credit card offered smaller lines of credit to women than to men, despite gender not being part of their algorithm. Apple’s inability to explain this was widely reported in the press and became a public relations nightmare. This example demonstrates the risk of consumer-facing AI applications from a PR standpoint and  is another reason why retailers may not implement them. 

Concerns about appropriateness. Sometimes AI applications are perfectly legal, but these applications can still spark controversy and public backlash. For example, AI is increasingly used by retailers to monitor the facial expressions and track the mood of in-store customers. This data is then used for personalized pricing or in-store advertisements, a strategy  about which many customers, especially older ones, have expressed concerns. Pam Dixon of the World Privacy Forum said the “creepy factor here is definitely 10 out of 10” and that retailers could potentially push certain medications to customers perceived as “sad.” As with privacy and bias concerns, the potential backlash from AI impacts consumer-facing applications more, making retailers more likely to adopt non-customer facing applications instead. 

Discussion and Conclusion 

The presence of artificial intelligence in the retail industry will likely grow in the years to come, making it increasingly important for retailers to understand these applications and how they can effectively implement them into their business. Specifically, the authors explored the differences between customer-facing and non-customer facing applications, finding that the risks for retailers are often greater with customer-facing AI, especially when the applications are in person or pose potential ethical concerns. Still, the authors remain “very (very)” optimistic about the impact of AI on retailing: “Not only are retailers expected to use AI to significantly add value to supply chain operations, but also retailers can use AI to suitably analyze the significant amounts of customer data to deliver high-value recommendations. Retailers who can suitably harness the power of AI will thrive.”

Hyunseok JungHyunseok Jung is an assistant professor in the department of economics at the Sam M. Walton college of business. He is an econometrician working on various topics: Big-data/Machine learning algorithms, network/spatial models, and firm-level productivity analysis. His research has been published in Journal of Business & Economic Statistics, Journal of Productivity Analysis and Journal of Retailing. 

Lucas Cuni-MertzLucas Cuni-Mertz is a recent graduate of the University of Missouri-Kansas City, where he majored in communication studies and political science. Lucas served as a newsroom intern at KCUR 89.3, as well as the editor-in-chief of the UMKC's student newspaper, Roo News. He currently works as a freelance writer and editor in Kansas City.