Nordstrom was an early adopter of all things digital and its approach to data science shows how the Seattle-based retailer is staying on the frontlines of using numbers, models and algorithms to deliver the customer experience on which it stakes its reputation.
As Nordstrom’s manager of customer and product data science, Anne Lifton is involved with many of the behind-the-scenes projects that improve the customer’s experience browsing and buying online. Retailers trying to keep up with consumer expectations of seamless interactions and commerce touchpoints across devices and platforms are starting to incorporate microservices—a way of developing software that helps them “adapt to new innovations more quickly and choose best-in-breed solutions,” per a BigCommerce blog—to turn on new features and capabilities.
But Lifton offered a “big life hack” for anyone managing data science for their company. The most meaningful microservices don’t solve just one problem, she said Tuesday at the Ai4 Retail conference in New York City, but should kill two (or more) birds with one software stone, especially when considering who the end user is, why they’re using that piece of software and what they’re trying to achieve.
With data continuously pouring in—every e-commerce interaction, every transaction is a new piece of data—retailers can get bogged down in figuring out what it all means, and how best to react.
There’s value in understanding the brands that trend quickly with customers, Lifton said, as well as those that might take a little longer to hit their stride with shoppers. Parsing the differences between one versus the other can steer retailers in the right direction if they’re trying to promote a label or serve up recommendations.
For example, the fact that someone with a long-running interest in Nike decides to shop Reebok products for a few hours one day doesn’t negate her ongoing affinity for the Oregon-based athletic brand, Lifton said. It’s the totality of the data, and perhaps the number of interactions with Nike products over time versus those from Reebok, that paint the clearest picture of what’s most likely to resonate with that customer.
Some retailers see true personalization as sort of the Holy Grail in e-commerce, and Zulily’s 1 million tailored homepages customized down to whoever’s logging on often comes up in this customer-centric conversation. Offering up seven-digit options for just the homepage requires rigorous A/B testing early on, and “those things take some pretty adaptable microservices,” Lifton explained.
“If you can tune how the API works in real time, now you’re five steps ahead of the game because people are just able to interact with the website and get all this cool feedback…that’s happening,” she said. “That’s the kind of flexibility that the future…could entail.”
Recommendation engines drive tangible value for online retailers, especially in apparel, but Lifton believes recommender systems that spit out suggestions in real time “will be an interesting change in the world,” but also the “biggest engineering challenge I’ve seen in the data science world.”
Nordstrom’s “looks” feature uses professional stylists to assemble fully curated outfits. Because any given shirt might be an element of two dozen stylist-created looks, the challenge lies in identifying the one ensemble that best matches an individual customer’s particular taste and style, Lifton noted. Today, Nordstrom uses a collaborative filter running in batch mode to achieve the styling functionality online, and though it’s effective enough, building outfits on the fly in real-time with up-to-the-minute information could take the experience to the next level.
“There are companies that are solving [this] for us and selling us, but I’m always a little bit hesitant,” she said. “We have such a build culture in Seattle. So I’m always hesitant of companies selling models.”