Stephanie Cutter |
On Jan. 22, [2012,] a young woman in a socially conservative corner of southwestern Ohio received a blast email from Stephanie Cutter, a deputy campaign manager for Barack Obama. Years earlier, the young woman had registered for updates on Obama’s website[. Then] she received from Obama and his Organizing for America apparatus [email] appeals to give money and sign petitions.
Cutter’s note was different. She boasted of a new administration rule that would require insurance plans to fully cover contraception as part of the president’s health care reform law, and encouraged her recipients to see the policy as reason to rally around Obama’s re-election. “Think about how different that is from what the candidates on the other side would do,” Cutter wrote. “Our opponents have been waging a war on women’s health—attempting to defund Planned Parenthood, overturn Roe v. Wade, and everything in between.”
It was a message that sat well with the young Ohioan who received it. She was single, liberal, sensitive to medical costs—but she had never told the campaign any of those things, and the one piece of information she had provided (her ZIP code) could easily mark her as the type of traditionalist Midwestern woman who would recoil at efforts to liberalize access to birth control. Indeed, she found it hard to believe that many other residents of her ZIP code would look as favorably upon a rallying cry to defend Planned Parenthood as she did.Issenberg reported just how Obama’s project--code-named Narwhal (a predator white whale only found buried under the polar ice cap)--linked completely separate sources of information so that every fact gathered about a voter became available to every part of the campaign. That meant the person crafting a provocative email about contraception would reach women that data-mining targeters had already pinpointed as likely supporting Obama’s views.
Already by early 2012, Issenberg realized Obama’s team could target online communication as sharply as had traditional human canvassers. It used extensive predictive models of individual attitudes and preferences to find friendly recipients. In the Cutter blast mentioned above, the campaign only pulled email addresses of women with flags marking them as likely pro-abortion rights. Narwhal also helped canvassers knock only on doors of known Obama supporters, and if a donor had given the maximum permitted $2,500, emails would stop requesting money and ask for volunteer time instead.
Rob Bluey leads the digital-media team at the Heritage Foundation. Shortly after the election in the Atlantic, Bluey wrote about another face of the Obama high-tech operation, “the Cave.” As with the movie “Moneyball,” the Obama folks focused on data that would give them an edge. The campaign left nothing to chance. Each night in the final stretch of the race, Obama's “Cave” analytics team ran 66,000 simulations through its computers to have a fresh perspective on the battleground states.
That real-time data then drove decisions on how to spend money to greatest effect. As Obama's campaign manager Jim Messina said of his campaign, "We were going to demand data on everything, we were going to measure everything. We were going to put an analytics team inside of us to study us the entire time to make sure we were being smart about things."
The Obama campaign excelled at using emotion to persuade voters -- ironically while relying on team of self-described nerds in their Chicago headquarters who helped optimize emails, build polling models, unfold a communications strategy, and create a social-media army, using analytics to make the campaign better.
Bluey tells us that after the campaign ended, Patrick Ruffini and his team at “Engage” put together a slideshow about Obama’s nerds titled INSIDE THE CAVE (PDF here). What follows is from Ruffini’s slide show.
“The Cave” was home to the campaign’s Analytics team, where behind closed doors, more than 50 data analysts used Big Data to predict the individual behavior of tens of millions of American voters. Despite being evenly matched financially, Obama for America (OFA) conceived of and built an operation 4 times the size of its competition:
OFA didn’t hire your typical political staffer. They went directly to Silicon Valley, to Fortune 500 data analysts, and to academia.
The Analytics team used its data to allocate resources in real-time:
• In Ohio, OFA had ballot test data on 29,000 voters, more than 1% of the electorate, allowing for deep demographic analysis.
• The final simulations were accurate to within 0.2% in Ohio and 0.4% in Florida, but were 1% too cautious in Colorado.
OFA measured the electorate not with a national poll, but rather with a “Battleground States Survey”--a single poll taken across battlegrounds (CO, FL, IA, MI, NV, NH, NC, OH, PA, VA, WI) initially once every three weeks, then 2 per week in final 2 months. They used “State Tracking Polls,” three-day rolling samples of 500-900 voters in each state. And the Cave’s Analytics Department had its own live callers making 8,000 - 9,000 calls per night, providing large sample sizes with short questionnaires.
Cell phone use changed everything. With cell phone-only households approaching a third of the electorate, traditional polling began to fail. The final “RealClearPolitics” average showed Obama up only +0.7%, but the actual result was Obama +3.9%, an error influenced by missed cellphones.
Traditionally, "voter identification" means contacting most or all individual voters in a state to identify "get out the vote" targets. The Obama campaign's approach was different: they called very large random samples of voters to develop statistical models that generated scores applied to all voters, then used in get-out-the-vote and persuasion targeting. Here’s an example:
In late October, Hamilton County (Cincinnati) releases the names of 103,508 people who voted early. Armed with Obama support scores for every voter in Ohio, Analytics director Dan Wagner matches these voters to the model. 58,379 of them have Obama support scores of 50.1 or more, for a projected raw vote lead of 13,249 in the county.Obama’s support models could change to reflect shifts in public opinion. OFA also had a “persuadability” score that modeled how susceptible an individual was to changing his or her mind based on campaign appeals. OFA built a tool for looking at the coverage of speeches in local newspapers so it could break down by geographic region how people reacted and which parts were quoted most. Speechwriters then knew how to tailor messages to convey what they actually wanted to get across.
OFA turned around initially weak fundraising and raised $1 billion by sending a lot more email than 2008 (404 national fundraising e-mails in 2012); it tested everything, and; it made people think they were going to lose. From the tests, “We basically found our guts were worthless." Ugly stuff won, like emails with yellow highlighting. “Quick Donate” (credit card info stored for one-hit; app on mobile) yielded 1.5 million users and $75 million that would not have been raised without the program. Donors gave 4 times as often and gave 3 times as much.
Tests on their donation page resulted in a 49% increase in their conversion rate. And by making the donation platform 60% faster, they drew a 14% increase in donations (speed matters). They found “you can get more users to the top of the mountain if you show them a gradual incline instead of a steep slope.” The end result of all their work was:
The Cave developed a “Call Tool” that allowed supporters in non-battleground states to use their home phones to call voters in battleground states.
Even though Hurricane Sandy caused critical infrastructure to go offline, because OFA had prepared for the worst, the team was able to keep its site up.
What else did the Cave yield?
• 1.2M active Facebook app users
• Use of a single app throughout the campaign, allowing the Cave to build up a massive install base and add features down the road
• 34M+ fans on Facebook
• 98% of the U.S. Facebook population were friends with someone who liked Barack Obama
• 24M+ followers on Twitter 30-40 tweets from @BarackObama daily
All this accomplished with an incredibly lean social team of 4 people.
OFA (with the Democratic National Committee) became the first political campaign in history to spend more than $100 million on online advertising, with digital's share of the media budget 21%. Online Ads drew direct responses, with ads featuring Michelle Obama performing best. OFA got more money by asking people to sign up first then immediately skipping those people to a donation page, not by starting with a donation page.
How Did OFA Innovate?
How did OFA reach youth?
Approaching the 18-29 year old demographic, OFA came to the startling discovery that while 50% of their targets were unreachable by phone, 85% were friends with an Obama 2012 Facebook app user. So they launched “targeted sharing” to Facebook friends who were voters in swing states. Like Quick Donate, integration with the rest of the technology stack was key. Users received an email requesting that they contact six specific friends, providing them their names and photos. As a result, 600,000 people reached 5 million voters, and 20% of those 5 million took some action, such as registering.
How did OFA buy TV?
OFA built a technology known as Optimizer to buy television ads in the same way that you should buy online ads, by focusing on audiences, not channels. It collected data about supporters’ TV viewing habits in coordination with a company called Rentrak; then for each channel and time in a swing state, they projected how many targeted voters were watching specific shows at a specific time, and; they went after the shows deemed most cost-effective.
OFA believes Optimizer made the TV buy as a whole 10-20% more efficient--the equivalent of $40 million to $80 million in added media. Optimizer meant buying micro-audiences no one else would think to buy, spending less per ad. At one point, the Obama campaign was up on 60 different channels compared with the Romney campaign's 18 during the same time period.
Real-Time Analytics Overtakes Polling
The Cave’s trend towards real-time analytics, and towards treating voters as individuals rather than as members of crude subgroups within a poll sample, should continue to evolve. Future campaigns will better understand and model the relationship between online conversation and public opinion. In four years, the media will stand up their own “Caves” to better understand how voters are moving in real time. And in eight years, “We will have difficulty telling a Field Director apart from a Digital Organizing Director. They are one and the same in future campaigns.”