The political world can’t seem to stop talking about data-driven voter contact, and for good reason—it seems to work. How to make it work for your campaign is a big question, but fortunately for us, it’s one that 2012 veterans can talk about now that their campaigns are over.
For example, the Data-Crunched Democracy conference attracted a swath of academics and political practitioners to Philadelphia in May with fascinating results (for one thing, it became obvious that the academic and political tribes don’t seem to speak a common language). Most valuable for our purposes: several of the top Obama 2012 data folks were in the room, and they helped fill in plenty of details about the most data-driven presidential campaign in history.
Better than random
One common theme? The picture often portrayed—the Obama campaign as a relentlessly efficient data juggernaut—paints over a lot of workarounds, hacks and improvisations. I’d heard this before, for example at CampaignTech in April, when Obama data manager Ethan Roeder had mentioned that plenty of the campaign’s technology was held together with “duct tape and baling wire.”
He echoed that sentiment in Philadelphia, and he wasn’t alone: Obama Chief Scientist Rayid Ghani said that for every mention of “data integration” on the campaign, he had “20 caveats” about how less-than-perfect it actually was in practice.
Sometimes the integration problems had to do with the speed of technology development, since the campaign had to simultaneously build, deploy and update its data platforms. It’s a bit like building an airplane in fight. Other times, integration speed bumps were inherent in the information itself. For instance, Facebook data could only be used within Facebook, not exported to the campaign’s other databases. The campaign could use data in Facebook via uploading a list of email addresses to create a “custom audience” for ad targeting. The campaign could also access users’ Facebook friend lists through technology similar to the NGP VAN “social organizing” tool we covered in TechBytes last year.
When supporters installed the campaign’s recruit-your-friends Facebook app, they gave the campaign’s data-management system access to their Facebook relationships, and the technology in turn used that information to ask them to reach out to high-priority voters among their friends.
The campaign had no phone number for roughly half of younger voters, and it relied heavily on Facebook-driven peer-to-peer outreach to connect with people under 30 (about a million people installed the app, yielding ~5 million voter contacts).
Despite the usefulness of Facebook data, it lived in its own universe: only if a targeted voter voluntarily interacted with the campaign would his or her information begin to flow into the data coffers. In fact, Ghani said that the Obama voter file was actually the smallest data set he’d worked with as a technology professional, in part because people vote so rarely.
Elections simply don’t come around that often, and compared with commercial marketers (who can draw on thousands of purchases and other transactions to predict buying patterns), political campaigners don’t have much historical data to work with.
Ghani also used the statement “better than random” to describe the campaign’s hopes for political data, i.e., that it would help the campaign allocate resources better than randomly throwing them at a map. Of course, I suspect they achieved results significantly better than random, but his words are a useful antidote to some of the more breathless coverage of the political data world we’ve seen lately.
The upshot: while the Obama data operation was unprecedented in scale and impressive in effectiveness, it wasn’t perfect. Online politicos feeling utterly unworthy, you can relax—a little. They were human, too.
More on data-driven TV buying
Also at the political data conference: Carol Davidsen, who ran that Obama campaign data-driven media buying effort we covered a couple of issues back. Interestingly, Carol’s background was in TV ad targeting based on set-top box data, not politics, and she and her team bought television time in a way that future campaigns should study.
She wasn’t concerned with the ad creative; her job was to reach the right people (specifically: the right people’s televisions when they were on and presumably being watched) at the most cost-effective times of day. The tools: a complex picture of the various voters the campaign needed to reach, used against data from the set-top box providers about the demographics of their customers and who was watching what when.
Carol noted that while political advertisers have grown comfortable with certain ways of doing business, media consumption habits have changed: roughly 40 percent of us no longer watch local news. Her job was to catch viewers in the exact hours when they were watching TV, something that’s obviously easier with someone who watches 10 hours per day rather than someone who watches two.
And it worked—Obama got in the neighborhood of twice as much ad time for every dollar he spent compared with Romney (though I’ve also heard that’s not a typical result to expect). Watch for more developments on this front.
Facebook custom audiences + lookalike targeting = wow
Let’s circle back to Facebook Custom Audiences, which allow advertisers to upload a list of emails that Facebook then matches with its users for ad targeting. Campaigns can use custom audiences to target their supporters with GOTV ads, for instance, or to exclude their existing base from recruiting ads.
Particularly powerful: combining a custom audience with “lookalike” targeting by uploading a list (perhaps a segment of an email-appended voter file) and then targeting ads at people demographically similar to the list. Via Facebook Exchange, ad vendors can do something similar with a cookie-targeting pool.
Better data-driven Facebook targeting is another big trend to watch.
Colin Delany is founder and editor of the award-winning Epolitics.com, and a 15-year veteran of online politics. See something interesting? Send him a pitch at firstname.lastname@example.org