What Online Retailers Got Wrong about Algorithms and AI
Around the time that the COVID-19 pandemic took hold in 2020, a group of ecommerce, direct-to-consumer fashion, personal care, and prepared meal-kit companies were being hailed as leading-edge retailers reinventing the online shopping experience by crunching data on customer behavior.
In 2018, industry trade journal RetailDive.com declared Katrina Lake “Disruptor Of The Year” for her role as founder and CEO of Stitch Fix, a fashion site offering a subscription service of goods curated by 3,900 part-time stylists. In an article published in the Harvard Business Review around the same time, Lake described her company as “a data science operation,” with revenue “dependent on great recommendations from its algorithm.”
Stitch Fix has been among the more visible examples of the rise of so-called subscription box retailers. The list includes beauty products retailer Birchbox, which “curates” and ships to subscribers a collection of products based on previous purchases and algorithms that categorize consumers based on age, location, and other data points. Blue Apron, a prepared meal subscription service, was another notable entrant.
At the beginning of 2021, three years after the company went public, Stitch Fix’s market capitalization was a whopping $10 billion.
Today, just eighteen months later, the stock has lost about 95% of its value and the company is expected to post its first annual sales decline since it went public in 2017.
Similarly, Blue Apron has turned into an even uglier investment train wreck — five years after its stock debuted at $140 a share it is trading at less than $4.
Why did the disruptors get disrupted?
As it turns out, the warning signs were clear back in 2018. In a piece that appeared on Quartz.com, Luis Perez-Breva, a lecturer and a research scientist at MIT’s School of Engineering, warned that, “Many retailers have forgotten what really helps customers: In-store assistance from human workers.”
According to Perez-Breva, “In order to receive clean data for machine learning (Artificial Intelligence or AI), for instance, many retailers send customers questionnaires which are easier for computers to process.”
But, he says, “Customers aren’t AIs. Most never answer the questionnaires, and many fill in whatever they remember. This leaves retailers with faulty … data.”
Also in 2018, consulting giant McKinsey & Co. surveyed more than 5,000 US consumers about subscription services and found that, “churn rates are high (nearly 40 percent) … and consumers quickly cancel services that don’t deliver superior end-to-end experiences.”
The McKinsey report concluded that, “Consumers do not have an inherent love of subscriptions. If anything, the requirement to sign up for a recurring one dampens demand and makes it harder to acquire customers.”
Meanwhile, several academics have written about the risks associated with collecting data on individual shoppers. It may be helpful to a consumer that a retailer knows their shoe size and favorite color. But what happens when the data collected by AI and algorithms includes the purchase of birth control pills?
To a longtime participant in and observer of the retail industry, an old maxim comes to mind: the more things change, the more they stay the same. AI is a powerful tool in the management of logistics, inventory, and a host of other business management concerns. In the case of anticipating consumer behavior, some of it is valuable but only if used properly.
If retailers want to know what consumers want, they have a time-tested way to find out — by consumer testing products and prices before committing precious capital. Instead of crunching data based on past behavior, or “curating” the profiles of consumer subgroups based on machine learning, retailers can more accurately predict trends and future demand by using real intelligence gathered from real-time online with real shoppers. And, if you are going to apply an algorithm, you better be able to prove it works time and again.