The eBay auction site gets to collect reams of site visitor data, just by its nature as an auction site. Visitors express what they're interested in buying. Through their bids they indicate their price range. Then, purchase decisions separate out the window shoppers from the actual buyers.
Alex Weinstein, director of marketing technologies, for the last two years has been taking that rough assortment of data and adding another element: data captured from eBay's own email marketing campaigns when it sends out notices of what's available to those it thinks are interested. Lately, that data capture has had to take on a real-time quality.
Rudimentary analytics would allow eBay to send email campaigns on watch or purse sales to previous purchasers of those items. But Weinstein has no interest in sticking with the basics. He thinks a lot more insightful marketing can be done with the data he already has in hand, never mind what might be possible one day.
Not so long ago, eBay conducted its email marketing campaigns by going through its stored up lists of site visitors, looking for those who might be a candidate to buy a particular item or category of items that it wished to feature as being available. An eBay campaign today might easily consist of a list with a million names. Generating such a list by hand would be time consuming.
"We had to invest a lot of manual effort into writing the right queries and building the audience list. It was very labor intensive, and targeting wasn’t precise enough," said Weinstein in an interview.
Want to see how eBay has made use of machine learning? See What eBay's Machine-Learnng Advances Can Teach IT Professionals.
That effort has been replaced by a CRM system built atop eBay's mail system, SendGrid. The custom system routinely collects data out of visitors' interactions with the eBay site. When it comes time to address a potential customer in an email campaign, SendGrid allows eBay to deliver its marketing message in a format that includes 6-to-8 "slots" or small content windows that make up the message or accompany a core campaign message. The slots contain personalized offers to the recipient.
The offers are not like display ads. Rather, eBay draws on what it already knows about the target recipient to come up with specific enticements. A person who's bought gardening tools, for example, might be guided to a blog that tells her how to attract butterflies to a garden. There might be a special going on for gardening gloves that the recipient would be informed of. Or perhaps the recipient had made a recent purchase for a trip to Ecuador, and wares from that country's artisans would be shown.
To individualize its marketing, eBay does the opposite of creating broad classes of buyers based on categories that allow a large number of targeted prospects to receive the same message at the same time. On the contrary, a one-million email campaign will have one million sets of decisions to make on which slot content to offer each recipient. In addition, the message goes out with the slots already filled with content. The decision of what goes into the slots depends on when the message is opened, not when the message is sent.
Things Change Fast
Weinstein explains that sales on eBay attract buyers, and by the time someone opens a message, offers that were set as the message goes out are probably going to be irrelevant if the message is opened by the recipient several hours later. Creating interest in a prospect that can't be fulfilled is a way to increase customer frustration, not satisfaction, Weinstein noted. So eBay's Target Engine, part of the CRM system, has to individualize the slots in near real time as the message is opened.
"Our deals get sold out quickly… We have to do 'open-time personalization and rendering.' We select the most relevant offers the instant the user opens the email. That puts significant requirements on our marketing pipeline," Weinstein noted. Then he added, "But it creates opportunities for us to personalize better."
Nowadays, if eBay were to conduct a sizeable campaign, it might send out the first 100,000 messages, observe the results, apply them to the next installment of the campaign, then send out another 100,000 in the next increment.
The SendMail CRM system that eBay has built can collect feedback on how many messages were opened. Were the subject line and header effective? Did customers spend time with the message? How many scrolled through it and clicked on the offer? And how many of those continued the process through purchase? And if the prospect didn't open the message within an assigned timeframe, that's toted up as a negative event. This data can be collected by SendGrid in "industry-standard" ways without intruding on the customer, said Weinstein.
"Engagement data is coming back to us in real time. We feed that data to a machine-learning model that uses it to personalize the next set of messages we send, adjusting recommendations based on that feedback," he said.
That might mean a particular coupon had gained traction with more users than expected. A particular blog reference was proving more popular than expected and driving value in the eyes of the recipients. A particular interest in the iPhone or other mobile device was yielding big results on a discounted product set.
"We're able to adjust our offers based on the real time data," he said.
While eBay right now is concentrating on improving its email marketing, there's no reason the personalization techniques wouldn't work in other channels, such as ads geared to individual visitors to partner or third-party web sites or ads placed in social media, based on the user.
Other companies are experimenting in the same way with their own personalization systems but they tend to be tailored to either email or display advertising or social media. EBay wants its SendMail CRM system to cut across the channels that are being used and individualize the response, regardless of venue.
"This area of personalization is very much an area of focus. We're just getting started with this… We're trying to invest in the data scientists who can continually improve our campaigns.
Weinstein called eBay's position of interacting with prospects and customers "a treasure trove of information" that the company was starting to take advantage of. If it exhausts its own personalization data, it can turn to third parties with their demographics, including regional location and income data, that will add layers of understanding on top of what eBay already has.
"We're just getting started on this journey. We will develop better modes of personalization," he predicted.Charles Babcock is an editor-at-large for InformationWeek and author of Management Strategies for the Cloud Revolution, a McGraw-Hill book. He is the former editor-in-chief of Digital News, former software editor of Computerworld and former technology editor of Interactive ... View Full Bio