For a company that’s all about the data, it could be said that Zulily was getting lost in the numbers. The Seattle online retailer has spent nearly a decade as the model for “discovery-based” shopping; its consumers, mainly tech-savvy moms, quickly took to Zulily’s regular flash sales and personalized experience to find family apparel, toys and home furnishings, and its success helped Zulily receive a $2.4 billion purchase price from QVC parent Liberty Interactive in 2015.
Because of the nature of its shopping experience — creating sudden demand and tempting shoppers to regularly refresh and check back (91 percent of its customers return and buy again) — Zulily is heavily dependent on interpreting terabytes of data very, very quickly.
“Data drives nearly everything we do,” says Gaurav Tandon, director of data science and machine learning for Zulily. “Data is driving our decision making not just by the day, but by the hour and minute. We have huge marketing and merchandising teams that need accurate information about what consumers want at that moment, and we have to figure out how to get them that information.”
KNOWING THE CUSTOMER
The challenge faced by Zulily is similar to those faced by other retailers: the need for more sophisticated data analytics, and the corresponding problem in that data is spread among various silos and legacy systems, making the information hard to access, analyze and use.
“In order to transform the way they operate to meet increasing consumer expectations for engaging experiences, retailers are turning to providers like Google Cloud to build a unified layer of data,” says Kris Baritt, technical director in the office of the CTO with Google Cloud.
Zulily’s personalization model means a focus on creating one-on-one experiences with each customer. A shopper who has bought clothing in a particular style is likely to buy other pieces in the same style in the future. The goal is a platform that knows the customer and their tastes and presents them with merchandise they will purchase.
“We say we have 1 million versions of our site daily, and that’s not too far from the truth,” Tandon says. “We have to present what’s relevant for each potential customer, so we’re dependent on high-level machine learning.”
In 2014 Tandon’s team began to see what other large online retailers were finding, which was that machine learning and analytics were about to take a huge leap as cloud-based platforms heralded a new era for ecommerce. To keep up with expected growth in its customer base, Zulily needed to explore how to increase scale.
The company had hosted its data warehouse and Hadoop-based big-data platform on bare-metal servers and was finding that Zulily’s growth was becoming too much for the system. It also was unable to get the sophisticated data analytics it would need to continue to move forward.
After looking at a wide variety of platforms, Tandon felt the best fit would be Google Cloud. “There are many good services out there, but we felt our greatest need was flexibility,” he says. “Based on our reliance on big data and what we may need in the future, we saw that Google was a good match.”
The company’s business model has been dependent on the analysis of mountains of data in milliseconds. “We’ve always looked ahead to see that machine learning would drive what we do better,” Tandon says. “The Zulily experience works best when the platform is making product suggestions for a particular consumer in seconds. That’s not something the human brain can do effectively for millions of visitors per day.”
About 70 percent of Zulily’s customers are purchasing on mobile devices, so the company needed a platform that was fully mobile compatible. “That number will only increase as customers become more comfortable making mobile purchases and platforms become more seamless,” Tandon says.
The nuts and bolts of migrating Zulily’s database and systems into the Google Cloud took a great deal of thought. The goal was to create the easiest transition possible, one that would quickly show results to non-tech analysts across the company including merchandising, marketing and operations. Ideally it would make the platform easy to access and digest for everyone so that no one needed a degree in IT or statistics to figure out how to use it.
Zulily uses more than 500 “merchants,” staff who interact closely with product suppliers to figure out the best way to merchandise products and manage inventory. One of the goals achieved by the new partnership with Google Cloud was an ability to deliver information to the merchandising arm of the company. Over the two years that the platform has fully been in place, Zulily has increased its daily data collection points from 50 million to more than 5 billion.
This massive increase provides a wealth of nearly instantaneous analytics, giving the company a real-time, 360-degree view of an offering just minutes after it’s put up for sale on the website. Zulily merchants can quickly add inventory on items they’re responsible for and watch how different types of sales messages affect customers.
The platform also carefully analyzes clickstream data from customers. If a potential buyer appears to be lingering on a product page thinking about the purchase, the system sends a notice showing how much inventory remains. It can also push targeted offers to customers based on their previous buying habits.
One of the key elements Zulily wanted to see was how the Google system would interact with the company’s advertising platforms. The company generates 30-50 ads into Facebook automatically in real time to match its sales and at scale covering millions of potential customers. Tandon found the increased speed of the system allowed Zulily to take advantage of “in the moment” ad opportunities, getting just enough visibility to lure a customer to the site. The result has been a 20 percent improvement in advertising efficiency.
Another big movement in social advertising is in personalization, which has become more targeted with Zulily’s increased use of analytics.
In addition to leveraging personalization from a social ad perspective, Zulily used it in its mobile push strategy as well. “In 2017 we shifted the content in mobile push notifications to now include one-to-one content personalization and local send time changes,” Tandon says. “With the changes, we’ve been able to drive demand from mobile push notifications by more than 49 percent year-to-year in the first quarter of this year.”
Looking ahead, Zulily hopes its data-first strategy will allow it to continue to be a leader in the “discovery shopping” realm. “We think this style of shopping is entertaining for the customer and there’s a great deal of growth ahead for it,” Tandon says. “We’re looking forward to what the data will show for us in the future.”
John Morell is a Los Angeles-based writer who has covered retail and business topics for a number of publications around the world.