Pandora Looks Past the Tracking Cookie By Mining User Data
Pandora wants advertisers to know: No cookies, no problem.
Even if privacy concerns and a shift to mobile devices neuter the web's "cookie" tracking software, companies like Pandora can still target consumers with the data they get every day when users sign in.
Now Pandora is pitching ad buyers on two audience segments it's assembled exactly that way, one for Hispanic listeners and another for Spanish speakers in particular.
The Internet radio company -- which counts more than 200 million registered users and 70.9 million monthly active listeners -- stocks up on first-party data by collecting an individual's age, gender and zip code information when someone signs up for the service.
Recently Pandora began digging into that registered user data to mine specific audience segments that are similar to the cookie-based ones to which online display advertisers have grown accustomed to.
Late last month Pandora finalized the first two of these proprietary audience segments -- one for Hispanic listeners and a sub-segment of Spanish-speaking listeners -- and has been pitching these audience targets to agencies over the last few weeks. The initial ads aimed at those listeners will roll out "as soon as we get [deals] signed," said Pandora's director of product management Jack Krawczyk.
The new targeting options are "a lot more robust than what we see on the [AM/FM radio] side. It's pretty significant what they're doing," said Horizon Media senior VP and managing director of audio and promotions Lauren Russo.
Previously to target specific audience segments like Hispanic or affluent listeners on Pandora advertisers had to rely on cookie-based data collected by third-party companies. However, unlike the new segments based on Pandora's first-party data, those third-party cookies didn't work for ad targeting in Pandora's mobile apps and were used mostly for display ads, Mr. Krawczyk said. Of the $128.5 million Pandora recorded in second-quarter advertising revenue, 70% came from mobile, and audio ads eclipsed display ads by accounting for 60% of ad revenue.
Cookie-based targeting is also problematic because it isn't always accurate. For example, using cookies to track someone's web browsing behavior could infer that someone visiting sites for The New Yorker, Wall Street Journal and New York Post is an affluent New Yorker. But maybe that person is a well-read but broke New York University student.
Pandora's proprietary audience segments are similarly based on inference but substantiated by the service's first-party registered user data. To create the first two sub-categories, the company cross-referenced its registered user data with U.S. census data (i.e. publicly available first-party government data) to identify zip codes with high populations of Hispanic and Spanish-speaking people and ran tests overlaying the two data sets to infer which listeners fit into those buckets. Mr. Krawczyk said Pandora's proprietary Hispanic and Spanish-speaking listener segments proved to be at least 10% more accurate than those created based on third-party data.
Pandora aims to add two new segments every four to six weeks, depending on how quickly it can access and assess the data needed to model those listener groups, Mr. Krawczyk said. The next batch will surface listeners with high-household incomes through the combination of census data and registered user information.
Pandora doesn't plan to always limit itself to publicly available data to infer audience segments. To identify which listeners are most likely play video games, Pandora surveyed "tens of thousands of users," Mr Krawczyk said, and discovered that group to dovetail with men aged 18 to 34 years old who listen to electronic music.
Pandora is also looking into how it can derive such segments from its own data. In one example Mr. Krawczyk laid out, the company could peg people as parents based on their age and whether they listen to a children's music station and then factor in how long it has been since they last listened to that station to infer the child's age.
The immediate aim is to take advantage of Pandora's first-party data to better target ads on the streaming music service, but Mr. Krawcyzk didn't dismiss the possibility of using the data to target ads on others' sites or apps.
"It's reasonable to infer we would be interested in applying what we know about logged-in users across a network of services. However we wouldn't approach doing that until all the right privacy measures were in place," he said.