How Analytics Made Obama's Campaign Communications More Efficient
BY Sarah Lai Stirland | Monday, December 3 2012
Last Friday in Washington, D.C., both Evan Zasoski, Obama for America's deputy director for data production, and Michelangelo D'Agostino, the campaign's senior analyst for digital analytics, showed their progressive peers how analytics can make online communications far more efficient.
Instead of blindly sending out mass e-mails to anyone who signed up, they worked to build models to analyze those subscribers and their behaviors in order to increase the impact of the digital teams' work.
During a session last week, Zasoski and D'Agostino shared some of the kinds of tests that they did and models that they created -- although they cautioned that there was still a lot that they weren't free to disclose.
Here are a few of their points:
- There are different kinds of tests you can conduct on your e-mail campaigns: There's one-off testing, like the one that they did in June that raised $2.6 million, as detailed by Bloomberg BusinessWeek.
- Or there are "best practices" tests, where "you're trying to isolate one very quantifiable notion of optimization before going forward," said Zasoski. For example, the Obama campaign did a lot of best practice testing on how much money it should ask donors for. It looked to its historical data on those donors, and their patterns of giving to deduce how to approach the donor.
- The campaign examined donors' highest previous donation amounts and then did tests asking them for various percentages of those highest previous donation amounts. They found that all versions of those requests did better than simply plucking a random number like $25 and asking donors for that set amount.
- The campaign found that people "who tend to give a lot of money" are also very likely to be comfortable being asked for a higher percentage of their previous donation.
- Watch out for "Frankenstein sends:" Beware of combining subject lines and content that performed well individually on their own, and then assuming that the combination when put together will generate a strong response.
- Models based on recipients' past responses to past e-mail campaigns can help organizers more efficiently target their specific communications. These models can be used to winnow out e-mails for people who are unlikely to respond to a specific kind of request (like volunteering,) and keep them on lists where they are likely to respond (maybe they just prefer to donate online,) D'Agostino said.
- The Obama campaign's digital analytics department used R, the open source statistical programming language, to build its models.
- Use automation to score your models as much as possible.
- Give the end-users, the people sending out the e-mails, and so forth, direct access to the output of your models through Web forms, so they don't have to ask you repeatedly for access to the lists of e-mails, and you're not e-mailing around evolving and confusing spreadsheets of e-mails. For example, they should be able to use the forms to access the portion of the list that is most likely to go out and volunteer, and send messaging to that portion by pulling that list of people up through a Web form, D'Agostino said.
- The analytics department also used D3 for data visualization.