Alex Gage is a fidgety man. In the first five minutes visiting his office, you’re likely to see him slide back in his chair, run his hands through his thinning red hair, flip randomly through his 2004 Florida research, and then lean forward to make his next point. All the while, a steady breeze blows in from his office’s deck overlooking the Potomac River. It’s the luxe office you’d expect of the guy credited with inventing political microtargeting. And the restless spirit of someone who knows he’s got a lot of new competition.

“We’ve made most of the mistakes that can be made,” he says about the ongoing research at his firm, TargetPoint. That’s what it takes, though—a ton of trial and error. And Gage is determined to push the envelope in order to stay a few steps ahead of the young turks who all, now, say they do microtargeting. “Everybody’s microtargeting this or microtargeting that these days,” he says wryly. “When we started doing it, it was a series of analytical processes. Now people use the term in a way that is far beyond its simply being a technique. Now it means everything.”
 
The basics of what Gage and other pure microtargeters do is still the same. Survey several thousand people about their views. Add consumer data (purchased from corporations like InfoUSA and Experion) and proprietary information (gathered by the campaign or political party). Do some smart things with math to develop a profile of different groups of voters (aka clusters or segments). Then map those clusters back to the full voter file and, voila! You can suddenly see what sorts of voters are out there, what they probably think, which party they’ll probably vote for, and what messages may interest them. Easy, right?
 
Not so much. Especially if you’re looking for a revolutionary idea that will catapult a candidate far past 50.1 percent on Election Day. Since 2008 is the first U.S. presidential primary to use microtargeting, we decided to look for the “next big thing.” And the innovations we found might just change everything.
Real-Time Targeting and Influentials.

Gage’s firm is the granddaddy of them all for a reason. In 2004, TargetPoint’s data mining work in Ohio was largely credited with delivering the state—and the election—to George W. Bush. In the 2008 primaries, TargetPoint worked for Gov. Mitt Romney, whose 2002 gubernatorial campaign was the first field test for Gage’s targeting techniques.
 
The Republican firm is now devoted to finding a way to do real-time microtargeting.
 
“In 2002 and 2004, we took the data we had on June 1, we delivered the campaign a product on July 1, and they used that product from July 1 to Election Day,” says Alex Lundry, the firm’s research director. “It was never refreshed with new data from voter ID calls, door knocks and so on.”
 
Microtargeting 2.0, he says, will be when a campaign calls Jane Doe and everything she talks about is instantly uploaded into the microtargeting database.  Although these instant feedback loops are already standard in the business world, Lundry says they won’t be common in even presidential campaigns for several cycles. But TargetPoint recently took the first baby step with one of its political clients. “We’ve been working with something called microtracking, which is essentially applying microtargeting processes to a nightly tracking study,” says Lundry.
 
TargetPoint is also the only political firm using net promoter scores to find high-value voters known as “influentials.” Promoter scores are straight out of Harvard Business School, and represent how likely (on a zero to 10 scale) a person is to recommend that someone buy a given product.
 
“You capture an attitude that was previously hard to measure, like voter enthusiasm for a candidate,” Lundry says. “It’s also indicative of word-of-mouth buzz, peer-to-peer persuasion efforts, and so on. But then you can apply it to something like microtargeting, because it can help find influential groups.” Last summer, TargetPoint found a group of voters in Iowa likely to recommend Romney. The campaign says targeting that group—religious white males with a strong interest in technology—helped Romney win the Ames straw poll.
 
Motivationally Speaking
Most microtargeting models focus on issues and likelihood to vote Democrat or Republican. But that wasn’t necessarily going to help Bobby Jindal win his 2007 gubernatorial race in Louisiana, where most voters are registered Dems who typically vote Republican. Strategists realized that to avoid a run-off, Jindal needed 42 percent of culturally conservative Democrats to vote for him on Election Day.
 
So Republican Blaise Hazelwood, who heads Grassroots Targeting, built her model around that core group.
 
“Usually when you build models, you are building them on everyone [in a district],” she says. “But the cultural conservatives were our target universe. We actually did survey work and tracked them all summer long.” Using that research, Jindal talked to voters in each segment of this custom universe. And on Election Day, he hit the magical 42 percent and won the race. 
 
The flip side of talking to voters is talking to campaign staff. In 2004, few top campaign staff knew what to do with raw microtargeting data. The young nerds who had time to play with the data were the ones in the remote field offices. In 2008, they’re now running the show and want to be able to get their hands dirty, so to speak.
 
Grassroots Targeting is the first firm to create its own software so campaigns can do just that. It lets a campaign manager select the voters he’d like to reach—like married men who are regular churchgoers who make above $100,000. Since the software is web-enabled, direct mail and phone vendors can go in and use the data as well.
“If you’re spending this much time and money to put this together, people should actually use the data,” Hazelwood says. “I try to empower the campaigns as much as possible, because they know their campaigns the best.”
 
The Art of Persuasion
“When it comes to microtargeting, there’s all this hype out there, it’s all the rage right now,” says Brian Stults, a bit cynically.  A GOTV expert, he now heads data services for the nonpartisan research firm Polimetrix. “It’s great in the weeks leading up to the election where you’re trying to get supporters to the polls, but in the months prior to the election when you’re trying to convince people, you’re still sort of guessing like you were four years ago.”
 
His firm studies those elusive persuadable voters. A million people have volunteered to help, and every day Polimetrix asks at least 5,000 of them about things both political and nonpolitical. “Our premise is by watching how people’s opinions change over time, and learning about the drivers of those changes, we can learn about what types of people are susceptible to change and what kinds of stimuli can create that change.”
 
The firm launched the survey to help companies sell products, but realized they were on to something much bigger.
 
“If someone doesn’t change their preferences over time, those are the kinds of people campaigns don’t want to waste their resources targeting, because they’re not going to move their opinion,” he explains. By figuring out how to identify persuadables, their survey data has boosted the accuracy of political models up to 15 percentage points.
And while all microtargeters are working at the household level these days, Polimetrix has started looking at how the relationships within households affect persuadability. “If you are a Democratic female living with a Republican male, you may respond to a very different set of issues than a Republican female living with a Republican male,” Stults says.
 
Non-linear Clusters
Last summer, Amy Gershkoff reviewed all the targeting software on the market for her firm, MSHC Partners. She realized that corporate programs might work great for selling magazine subscriptions, but that they force voters into narrowly defined categories, like slicing a pie into wedges.
 
“In the commercial software, having big overlapping clusters is okay, because if you have a consumer who’s in two different clusters, you’re on the fence for sending them two different catalogs,” she says. “But in politics, you could wind up with people who have a 40 percent likelihood of being in a young conservative cluster, and a 40 percent likelihood of being in an older, blue collar, liberal-leaning cluster.”
 
Gershkoff’s A-ha moment came when she stopped trying to fit a round political peg into a square corporate hole. Her SmartClus software identifies people who cluster together naturally. So instead of a circle segmented into pie wedges, hers is broken into bunches of amorphous blobs—some big, some small, some containing smaller blobs of their own—that lasso together different groups of people. Voters who don’t fall into one of those blobs are set aside, rather than creating a less-accurate cluster. When she grabs my notebook and sketches a sample, the blobs look almost artistic, like a child learning how to draw circles.
 
“Most voters, male or female, are really complicated. So it makes sense to me that you would need something really complicated, really nonlinear to figure out how to talk with [them].”
 
Artificial Intelligence
In 2004, political consultants turned to the commercial world to see what great ideas they could steal. In 2008, the business world is reaching out to people like Ken Strasma, whose firm Strategic Telemetry is using artificial intelligence to build better, faster models for Sen. Barack Obama’s presidential bid.
 
There’s a scene in the 1980s flick Wargames where the computer, Joshua, plays tic-tac-toe. He gets faster as he figures out how to play the game. Everything starts whirring at light speed as his learning curve shortens. That’s basically what Strasma’s computers do. They race through thousands of options, learning from previous models they’ve already built, until they develop the best possible formula.
 
Strasma compares it to Darwinian evolution. “The model changes through survival of the fittest; the fittest organism survives to pass its genes on to the next generation,” he says.
 
The software first determines the “voter DNA”—the thousands of data points, anything from commute time to income to magazine subscriptions—that describes each voter in a survey. The computer then uses genetic algorithms to sift through each string, one by one, to find the sequence that best indicates a generic voter’s propensity to vote in a certain way.
 
He’s also experimenting with neural networking, which is another machine-learning algorithm. But instead of having a string that defines someone, it’s more akin to a complicated web of data that includes commute time, income, likelihood of being African-American, and so on.
 
It could take months to crunch through all the data, but Strasma’s found a way to use dynamic parallelism to make everything go faster. Sorry for the geekspeak, but dynamic parallelism means, basically, that all these computers can work on the same project at the same time.
 
The result is a stripped-down office loaded with flashy liquid-cooled computers. Oh, and a very understated bar chart comparing Obama’s Pennsylvania primary results to Strasma’s predictions. Those bars are close. Really, really close.
 
“Some of our scores for similar projects have been accurate within one-tenth of one percent,” he allows. Not shabby.
 
Attitudinal Data
The right message is easy to develop for a campaign’s core supporters. It’s harder for swing voters and others who may not, on first blush, share a party’s values. Those are the people who fascinate Bob Baskin. His start-up firm, Spotlight Analysis, has one focus: to get inside their guts.
 
“Everybody’s romanticizing about microtargeting, but they’re missing the point,” Baskin says. “There’s another level of microtargeting beyond consumer data. It’s a better way yet, and it’s called attitudinal data. It’s about how Americans really want to live their lives.”
 
Backed by Democrats like Herb Miller, Don Baer and Mike McCurry, Spotlight Analysis ran a $1 million national study of 5,000 people in 2006. It clustered voters into 10 segments based on their most deeply held beliefs.
 
Take the barnraisers, for example. They care about personal responsibility, integrity, about doing the right thing. For years, barnraisers voted Republican. But they were so horrified by that party’s ethics and sex scandals in 2006, Baskin says, that “my barnraisers came home and voted for the Democratic candidates in the 50 to 60 percent range.”  That margin handed Democrats their congressional majority.
 
So for one federal race Baskin is modeling this cycle, he’s advised the candidate that he has an opportunity to address the barnraisers by framing his opponent’s ties to big business in terms of fairness.
 
“We said, talk about the job losses these companies have caused—how fair is it that these companies are laying off your friends and neighbors but providing your opponent with large campaign contributions?” Baskin says.
 
Facial Recognition Software  
“Big Brother is here,” says data guru Bob Blaemire about the new facial recognition software from Microtarget Media. He means that in a good way. Since launching its political division earlier this year, the Canadian firm has campaigns on both sides of the aisle plenty excited about its facial recognition technology.
 
Let’s say John McCain holds a rally in Miami. As he’s running through his stump speech, talking about patriotism, helping small businesses and improving education opportunities, videocameras are filming the entire audience’s facial expressions. And let’s say one audience member, on her way into the rally, had signed up for a raffle and walked away with a keychain implanted with a small RFID (radio-frequency identification) chip. The campaign could use the chip to identify her, analyze her reactions to the speech, and send her a mail piece targeting the issues she cares about most.
 
"You put those together and you have the most powerful set of microtargeting tools that you could have,” says Terry Popowich, the firm’s president.
The technology can also analyze people’s emotive responses on topics where they’d be less forthright with a pollster.
 
“Imagine getting a focus group’s emotive responses to an ad before it runs,” Popowich says. “You don’t have to ask them about [a tough subject]. Getting people’s pure uncontrolled responses to an ad could be a very powerful way to understand what you need to do to elicit [geniuine] responses in people.”
 


These seven innovators, of course, are just the tip of the iceberg. Google is using geotargeting; eVoiceAmerica is inventing ways to get voters to volunteer personal data; Holinshed Research Group is using GPS to improve sampling quality.
 
Which is all to say that the innovations are only beginning. And the breakthroughs unfolding now will change the political terrain in ways we can’t fully predict. That’s probably why every firm out there wants to claim they “do microtargeting.” But the future belongs to the true pioneers. 

Christie Findlay is Editor-in-Chief of Politics magazine.