Without question, the pandemic has changed the way consumers shop. The number of online purchases has skyrocketed, with many people embracing the “click-and-collect” method. On the business side meanwhile, retailers are now expected to deliver services that are not only convenient and safe but also more personalized. To that point, it has been reported that 91% of consumers now prefer brands that offer recommendations and deals relevant to them as individuals.
Retailers have turned to algorithms to reach these pandemic-era demands. These programs analyze consumer data, generate key insights and even automate processes that can boost customer engagement, conversion and loyalty.
While algorithms help retailers increase efficiency, however, they’re also prone to bias that can exclude certain segments of customer bases. Below we’ll dive into the issue of algorithmic bias, how it can affect efficiency, and what this all means for the retail industry.
To understand this phenomenon, we first need to understand how algorithms work. Machine learning and deep learning algorithms are powered by AI, and they need to be trained with large datasets in order to generate insights and predict future trends. If you feed an algorithm millions of photos of a chair, for instance, it will be able to determine if images contain chairs or not. Algorithmic bias occurs when algorithms aren’t trained with sufficiently diverse datasets.
To give an example, Google Photos (rightfully) came under fire in 2015 after it tagged two Black people as gorillas. The unacceptable error was attributed to the fact that the app was not adequately trained to identify and tag people of color. Similarly, Amazon scrapped an AI recruitment tool in 2018 for discriminating against women applicants. The tech industry is mostly dominated by men, and as a result skewed datasets taught the algorithm to exclude candidates hailing from women’s colleges — and, in general, CVs that contained the word “women’s.”
As algorithms become staples in the retail industry, the risks of their developing biases remain. When bias occurs, even a best-case scenario will cause a business to lose customers. Consider for instance men who buy makeup to cover up tiredness, blemishes or other skin issues. If an algorithm uses purchasing history to make personalized recommendations, skincare brands may alienate such customers by sending offers for women-specific products.
The worst-case scenario, however, involves biases actively offending or discriminating against consumers, as in the examples mentioned previously. Case in point: retailers that use algorithms for inventory forecasting. A slump in sales won’t provide enough data to predict future demand, and so algorithms are likely to run into a higher margin of error. This can result in, say, clothing brands supplying more items for slim individuals than plus-sized persons. With consumers today expressing their preference for more inclusive brands, algorithmic bias may instead work against retailers.
Fortunately, there are many strategies retailers can employ to minimize algorithmic tendencies toward damaging exclusivity. The first is to be transparent with consumers. Disclose which of your services use algorithms and how these algorithms work. Next, always run algorithms under human supervision. Consistent monitoring serves as an added layer of quality control that can preserve fairness while detecting issues that need to be fixed.
Finally, be picky about which algorithms and providers you use. Consider companies that employ data scientists with advanced training in data analytics. Their expertise goes beyond simply generating key insights and visualizations from operational data. It also incorporates practical knowledge pulled from up-to-the-minute business headlines. Consequently, these professionals know how data can best optimize strategies and processes, forecast sales and market trends, as well as improve the overall customer experience — all in line with your existing business strategies. By working with companies that use diverse datasets for strategic training, you can ensure that only high-quality algorithms are employed in your operations.
So given the risks algorithmic bias poses to retailers, should they still be used? The short answer is yes! Algorithms in retail are ultimately more effective than exclusive. Since the consumer experience is now overwhelmingly digital, retailers need algorithms to leverage data and effectively engage consumers in this realm. They also need these programs to remain competitive. And finally, since bias can be reduced, the risks it poses can be made negligible provided the proper effort is made.
“It’s pretty obvious that most of the retailers that are in trouble failed to innovate,” SageBerry Consulting President Steve Dennis said in an exclusive interview. “Winning in today’s environment takes an attitude of radicalism and a willingness to embrace experimentation and a sense of urgency. As the saying goes, ‘The best time to plant a tree was 20 years ago. The second-best time is now.'”
Wise words to apply to the question of algorithms in retail, so long as the process is approached thoroughly and with care.
Jermaine Kaye is a freelance writer who specializes in pieces on lifestyle and tech. She’s particularly passionate about meditation, wellness and the many different ways businesses can use technology to stay competitive in the Digital Age. In her free time, she likes to read, paint, solve Rubik’s cubes and try out innovative vegan recipes in the kitchen. Once pandemic restrictions ease she hopes to continue participating in marathons and doing more in-depth research on the ever-expanding use cases of AI in the corporate setting. Moving forward, she hopes to use her growing portfolio to start her own ecommerce business.