How Granular Analytics Are Driving a More Sophisticated Approach to Campaigning

There are a lot of moving pieces, but the game’s objective is simple: Figure out which potential voters can swing the election, and fill their heads with bright ideas.

by / August 30, 2016

Elections are like wars. Though more civilized, an election is a conflict in the pursuit of power based on ethics like understanding the enemy, adapting to a chaotic environment, dissimulation, deception and gathering intelligence to inform resource allocation. The philosophies guiding the hand of a winning general haven’t changed since Chinese general Sun Tzu wrote The Art of War more than 2,500 years ago. The crucial difference is that the implements have changed.

Instead of gathering troops to intimidate the opposition, presidential candidates analyze statistical models and make bold claims about the loyalty of their Twitter followers. Instead of flanking the enemy, they formulate precise TV ad campaigns and bluff about which states they intend to contest. Tzu promoted the idea of fluidity — that a wise general adapts to his environment like water flows around rocks. Failure to adhere to this piece of ancient Chinese wisdom has already claimed at least one casualty in failed Republican candidate Jeb Bush.

Bush spent about $60 million on television advertisements, according to Prosper Insights & Analytics, but it got him no closer to the election. How people get their information, form opinions and interact with one another evolves daily, but a lot of politicians are acting like it’s still 1972, said Philip Rist, executive vice president of strategy at Prosper.

“For folks like us who spend time analyzing human behavior, we know that the reason [Bush’s campaign] didn’t work is they were putting all their money in two forms of media that are not big with the younger people. They’re putting it mostly in television and direct mail,” Rist explained. “They go do the old things that they always have done and now they’re trying to figure out why it’s not working. The millennials aren’t watching TV.”

Next to baby boomers and the retired, millennials are the most important age group this election, according to a Prosper analysis. Even Donald Trump, who called data analytics “overrated,” began seeking help from analytics companies on how to target voter groups. It’s hard to overlook data, because it powers all the tools candidates use.

Combining data from all over to form predictive models and spot trends is the skeleton of this beast. The meat on the bones is field tools — things like voter contact organizers and digital engagement tools, which are used to raise funds and ultimately get people to vote. There are a lot of moving pieces for data scientists to contend with, but the game’s objective is simple: Figure out which potential voters can swing the election, and fill their heads with bright ideas.

Analytics companies, which tend to lend their services exclusively to one political party or the other, include names like NGP VAN (Democrat) and Cambridge Analytica (Republican). A new addition to the scene is a company called Timshel, founded by Michael Slaby, chief technology officer for Obama’s first presidential campaign. The basics of campaigning haven’t changed in the last 100 years, Slaby said — but new tools like social media and an interactive Web make it easier to participate.

“I think the metaphor that’s most valuable to think about is building a community,” he said. “A campaign is fundamentally about building a community of people prepared to vote for you.”

The days of blindly knocking on as many doors as possible are gone, Slaby said. Campaigns are mixing analytics and game theory to maximize resources spent on the middle and swing voters before the November deadline arrives.

“Where are those people? What media are they consuming? What messages are most likely to move them and inspire them to act?” said Slaby. “And how does that translate into electoral votes? [It’s about] being able to optimize resources like, ‘Should we be buying TV ads in Pennsylvania or Ohio?’ Because if you’re definitely going to win Pennsylvania, like, you’re 100 percent sure, then you should spend zero dollars there. That’s tough. How confident are you?”

Showing up at the podium and running down the bullet points isn’t good enough. When just a few fractions of a percentage point can make a difference, candidates have to get scientific. Cambridge Analytica surveys what voters care about, but it also figures out why voters care about the things they care about, said Molly Schweickert, the firm’s head of digital.

A political ad with a broad message about the economy, for instance, is less effective than a specific ad about the economy designed for a certain audience. A young white-collar voter might care about the economy because he’s looking to buy a house and start a family, while a middle-aged blue-collar voter who already has a family might have different concerns, and therefore different emotional reactions to the connotations of various economy-centric messages.

Analytics determine the message’s vehicle too. Not everyone uses social media. Of those who do, some don’t like getting their news there. Some might be willing to engage in political discussion with friends and family, or maybe some like reading news on Twitter, but not Facebook. These are the things that data companies are feeding their models to help politicians figure out where their efforts go furthest.

“It’s not just classifying people into these top-line groups based on what we know about their demographics, but digging into the behavioral signals they’re giving out,” Schweickert said. “In most instances, we are able to work at the individual level or the individual device level, not just the entire household. That is a tremendous leap forward. ... What we’re doing is bringing humans back into targeting. We’re not thinking of them as cookies or as an address. We’re thinking of the human behind that.”

Kim Wyman, secretary of state for Washington, is campaigning for re-election this year, and while she said her own campaign isn’t as technologically sophisticated as politicians seeking more high-profile offices in the state, she’s seen the game change a lot since starting her political career in 2001.

A Washington state-based technology consortium called TechRoanoke, for instance, today helps nonprofit organizations and conservative political causes use data to further their goals.

“You can download a walking list to your supporters and they can go out and knock on doors, and they really tied in a lot of analytical data so you know just who you’re talking to when you go to that door,” Wyman explained, noting that in years past, the information available was as simplistic as whether a given household leaned left or right politically. “... That’s going to change some of that grass-roots campaigning in really powerful ways. We’re just starting to tap into it and use it, but it has applications for doing direct mail pieces that can be very specific to voters in ways that I don’t think we’ve ever seen before.”

On the left, another Washington-state-based group called Fuse Washington is helping the presidential campaign with tailored messaging. By definition, being conservative means being slow to adapt new ideas and methods, and so the left’s willingness to embrace new technologies is proving an edge for the party.

“They’ve been really effective in tailoring that messaging on a specific candidate or race to very specific voters and talk about it in those terms that really matter to them,” Wyman said. “Quite frankly, that’s what I’ve seen in this state and in a number of races nationally, why the Democrats have had really good success.”

Politicians from all parties are abandoning conventional wisdom as they find that norms relied on as recently as two elections ago don’t hold anymore. The Party Decides, a 2008 book by four political scientists, outlined a well accepted theory that a given candidate’s nomination is a foregone conclusion early in the process. The theory goes that party actors — those most involved in the political process like volunteers and other politicians — tend to rally around an electable, mainstream choice. And then came Trump, a man who until 2016 had been little more than one of America’s favorite punchlines.

Today’s political campaigns use common, ready-made tools like Twitter, Facebook, WordPress, DoubleClick and Google Apps, but the custom data engines that drive how politicians spend their time and money are what distinguish the modern campaign. During Obama’s second run at the presidency, his campaign experimented with the Civis Media Optimizer, a tool that helped it figure out which TV shows undecided voters were watching. By piloting variations of a similar advertisement in various small TV markets, Obama’s researchers were also able to see which messages were more effective with certain demographics.

What was experimental in 2012 has become necessity in 2016. The success of the Optimizer in Obama’s second campaign prompted the birth of a company called Civis Analytics the following year. What’s different today, explained David Shor, a data scientist with Civis Analytics, is the quality of the data, but also the scalability of the technology. It took 55 people working on one campaign in 2012, he said, but today, six or seven people can do the same work on 60 races concurrently.

“The entire reason Civis was founded is that we’ve realized that the questions campaigns face are actually very similar to the questions that other organizations face,” said Shor. “We work with a wide variety of nonprofits and corporations because basically they have the same questions our political clients do, which is ‘Is what we’re doing working?’ ... You wouldn’t believe how many business meetings I’ve been in where C-suite executives say they want to treat their company like a political campaign.”

The thing that makes tools like the Optimizer relevant is that they are rooted in empiricism. In 2008, common sense told everyone that a person like Trump could never take a shot at the Oval Office, and common sense was wrong. A wise general is rooted in reality. He looks at the facts and plies his ethics against what is real, no matter how counterintuitive.

Peter Bouchard, the director of media science at Civis Analytics, explained that for a long time, people believed that the most effective slot for a campaign ad was during television news shows, because it was thought that news audiences were both suitably interested in politics and primed at that moment to receive a message.

“When we actually did randomized controlled experiments and built models, we found that, generally, people who are into news are not persuadable at all,” Bouchard said. “The TV show we found that had the most persuadable voters was Dog the Bounty Hunter, which ends up being a lot cheaper.”

The use of data is the biggest difference in today’s campaigns, said Stephanie Hannon, CTO for the Hillary Clinton camp.

“It’s the power of data to be hyper-targeted in all the different mechanisms you’re using to reach the voters and that includes advertising, that includes email, that includes the website, that includes how you talk to people at doors,” Hannon said. “Data and really smart, rigorous data science is changing the way we talk to voters, and the impact of that is people get more personalized and relevant messages to them in channels they care about, from people they care about.”

The shortcoming of these efforts is that they stop after the election ends, she said, but there’s value to society in developing these technologies consistently. It just makes it harder to build a viable business model over the long term.

“I wish more money was going into those spaces, because it’s important and there are big problems to solve,” Hannon added.

Looking for the latest gov tech news as it happens? Subscribe to GT newsletters.

Colin Wood former staff writer

Colin wrote for Government Technology from 2010 through most of 2016.

Platforms & Programs