When BJ’s Wholesale Club on Thursday (May 3) said that it would leverage artificial intelligence machine learning in its mobile app, it joined the crowded club of companies boasting machine-learning capabilities while remaining vague on the details.
But the 215-store chain — operating in Connecticut, Delaware, Florida, Georgia, Maine, Maryland, Massachusetts, New Hampshire, New Jersey, New York, North Carolina, South Carolina, Ohio, Pennsylvania, Rhode Island and Virginia — pledged to use machine learning to boost its CRM shopper profiles and to immediately apply it to change mobile responses.
“The new discover feature lets shoppers explore new products and easily swipe right to add to a wishlist or left to dismiss a product,” the chain said in one of the shortest news releases that retail has ever seen. “Using machine learning, the discover experience will be personalized to each user based on previous selections they’ve made through the swipe right or left process.”
Why do I find this so interesting? Because mobile retail has been impressively remiss in this area for far too long. Given the very nature of mobile interactions, shoppers share far more clues about likes and dislikes than they do with desktop interactions. A desktop interaction with a web page can capture completed actions (making a purchase, putting something in a cart and then removing it, clicking on a product page for more details, etc.), but mobile can track far more.
Aside from the swiping right and left that BJ touched on — which clearly adds actionable data to the shopper’s profile — there is the geolocation data specifying where the shopper is at the moment of the interaction. The system can track a daily commute and note, for example, that this shopper will often look at product pages on the train on the way home, will never look at all on the way in and will only complete purchases while at home. That would suggest that a pop-up offer to close a deal (say an extra 15% off for buying right now) is wasted when the shopper is on the train, but an excellent approach when at home. The app (like its desktop counterpart) also knows the times when this shopper is more likely to be open to closing a purchase.
Another key area that retailers need to explore is the age-old single view of the customer. In this instance, that means being aware when the customer moves from the mobile device to a desktop and then potentially inside a store. Is this shopper inclined to research on a mobile device and then purchase on a desktop? Does the shopper explore on the mobile device and then drive to your store? That can be tracked quite precisely, with the mobile geolocation data indicating that the shopper is approaching a store and then the store’s Wi-Fi connection confirming it.
Or is this shopper comfortable with tendering a purchase on a mobile device? If so, is the shopper using a password-memorizing app? Passwords are often the key obstacle with mobile purchase completions. That said, biometric login on an app (fingerprint or facial recognition typically) can make a big difference with making shoppers comfortable closing the deal on mobile.
Let’s drill down a bit into this swiping business. BJ’s statement said it will be a right swipe for a shopper to add an item to a wishlist and a left swipe to dismiss a product. What new data is this likely to deliver? The right-swipe effort won’t yield anything new, since shoppers already had a way (shopping cart or wishlist) to note what they wanted to pursue. It’s the left-swipe action that is potentially new data.
Unfortunately, this raises the key question: Why would a shopper bother? Let’s say that a customer is interested in a new air conditioner. She’ll likely scroll through all of the AC options until she finds one she likes and then act on that one. Why would she bother swiping left on something she doesn’t care about?
“This looks like a Tinder setup. But who among BJ’s customers would spend their time saying, ‘I like this and I don’t like this product?'” asked retail technology consultant Todd Michaud. “What problem for the customer is this solving? What about this is worthwhile for the customer to do? And without this providing benefits to the consumer, why on earth would they do this?”
Another consideration: This binary choice — like/dislike, right/left swipe — is not especially meaningful, Michaud said. Most products will be irrelevant, which is very different from being one the shopper doesn’t like. Consider, for example, a person who has no babies being offered a steep discount on diapers. Her accurate answer may be that she has no interest because she has no need. Forcing her to swipe left to indicate dislike will do little other than give your team misleading data.
Michaud added that there is a reason for the chain to want this. That involves the tendency of people to be more truthful when listing dislikes than listing likes. To be polite, someone might say they like something that they really don’t care about. But saying “dislike” often indicates a stronger sentiment. “So I get it in concept, but, again, why would customers do it?” he asked.
Here’s something else to consider and it involves machine learning and every IT person’s least-favorite subject this month: the European Union’s General Data Protection Regulation (GDPR) set to go into effect later this month. There is a GDPR provision about automated decision-making that says that “the data subject [consumer] shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.”
Does “significantly affect him or her” include offers and special discounts? If so, does this mean that BJ’s will need to get special permission from each mobile shopper before doing this with each one?
On BJ’s swiping and machine-learning effort, here’s my take: The swiping is a smokescreen when it comes to machine learning. BJ’s will undoubtedly use machine learning to analyze everything it can dealing with mobile’s attributes, in-store’s attributes and desktop interaction attributes. But there’s simply not likely to be enough new data from this swiping effort to make a difference. Hence, it wanted to tout the consumer-oriented swipe effort while somehow mentioning machine learning.