FORTUNE — A revelation that will shock corporate-types: Until very recently, political operatives didn’t bother applying even the most basic return-on-investment measures to their campaign spending. In other words, they had no idea if a volunteer knocking on doors was more valuable, in terms of real votes, than a multi-million dollar TV ad. But the science of campaigns is undergoing a revolution, vaulting the field to the cutting edge of data analysis and behavioral psychology, and remaking the way candidates hunt for votes. Sasha Issenberg details this change in his new book, The Victory Lab. Fortune spoke to Issenberg about what this might mean come November 6.
Why did it take so long for empiricism to penetrate campaigns?
Campaigns are like the worst corporations in America. They exist for six months or nine months to win market share on one random Tuesday in November. The day after, the corporation goes out of business, and everybody scatters. There’s no incentive to do even a basic post-mortem about what worked or what didn’t, because the candidate is either off trying to govern, or run a transition, or on vacation feeling sorry for himself and doesn’t want to revisit what went wrong. Nobody wants to invest in the research that will yield lessons after Election Day. So campaigns have been traditionally in hock to consultants and vendors, because they’re the ones who have permanence.
Either the CEO or the chairman of the campaign is a candidate who either has run for office before and thinks he’s figured it out and is going to do the same thing again, with the same people that they did it last time, or it’s someone who’s never run a campaign, so there’s no reason why somebody running for Congress for the first time should have any idea how you run a $10 million statewide marketing company for six months. They rely on consultants and vendors, who have an incentive to do the same thing over and over again.
Campaigns haven’t been built to want to invest in empirical methods, even if they existed elsewhere. The places where the innovations take place are institutions that break the cycle of yearlong campaigns. So they’re either in parties that make a serious investment in new tools or techniques, or institutions like the AFL-CIO [American Federation of Labor and Congress of Industrial Organizations] that are going to be around forever; or in consulting firms or vendors who are willing to make unusual investments in collaboration and learning even if it won’t yield short-term benefits; or presidential election campaigns, which exist for four years and can start to think in the way that a Fortune 500 company would think about the problems they’re trying to solve.
So that’s why they were slow, and it makes sense. Campaigns aren’t around long enough to want to do ambitious things in terms of method.
But the people who staff the campaigns don’t vanish. So it makes some sense that a group like the Analyst Institute — the six-year-old organization coordinating these efforts on the left that you write about in the book — would emerge.
It makes intuitive sense, but it also was a really radical proposition. At the beginning, a lot of the people participating in the group had polling firms and competed against one another. These were people who pitched the same campaigns or state parties for business. And the Analyst Institute created an environment where people went in and presented what could be competitive advantages to their competitors. Somebody does a test and comes up with a major breakthrough and then they run to show a PowerPoint presentation to people who they compete with. And what you see, I think, is how much people losing and being scared of the consequences of that are, are willing to change their behavior in ways that could be economically irrational. So the mythology of Rove that emerged after the 2004 campaign scared Democrats into collaborating in ways that weren’t necessarily in the interests of their short-term bottom-line, but were good for the party and the movement.
So what do they do now?
Now campaigns are using reams of new data from sources outside politics, statistical modeling, and in some cases ongoing randomized experiments to profile every voter in America, predict what he or she will do on election day, and isolate what exactly it takes to change their opinions or modify their behavior.
And you give Democrats a competitive edge over Republicans in pioneering and incorporating these methods?
I think it’s as big an edge as it’s probably ever been. I mean the gap in intellectual sophistication, in talent, and ambition is massive. The left is way ahead of where the right is. It could be temporary, but the left has done everything right to make it a structural advantage by building institutions like the Analyst Institute, that are more permanent than any single campaign will be. They’ve built both a culture and a set of institutions that are committed to R&D. They’ve done all the right things to break out of the cycle of doing the same thing every two years over and over again. So it’s a massive advantage.
And much like the Bush campaign in 2004, which benefitted from four years of laying out a deliberate research agenda and having control of a party to test new techniques and enforce them and invest in data analysis, the Obama campaign has benefited from having four years in control of the DNC to do that.
How’s Romney’s presidential campaign doing with this stuff? You talk in the book about how he was on the front edge of it in his 2002 gubernatorial bid. He’s also been running for president for, effectively, six years. We know he was heavily data-driven in his private equity life, too.
Since his 2002 campaign, Romney’s had a commitment to using data for targeting. In that race, he benefited from having a number of people with him who’d never worked in politics but came from the corporate world, and were willing to embrace new uses of data for campaigns. And he had this challenge in Massachusetts running as a Republican: He knew he had to pick off Democrats and Independents to put together a coalition. Any tool that would help him identify those people was of great use. His was the first campaign to use what we now call micro-targeting, and Romney probably had the most sophisticated data operation in the Republican primary in 2008, and certainly did again this year. His operatives showed themselves in the primary to be really competent and savvy at running a modern political campaign, which was not true for Gingrich or Santorum.
But they’ve been trying to build in six months what the Obama people have basically been building continuously for six years. And Romney’s cheap, which manifested itself in having a very small footprint during the primaries. There’s sort of a virtuous cycle the Obama people have perfected where having good targeting data is most useful if you have a lot of field activity. At its core, it’s best for sorting out which doors to knock on, and where to send mail, and who to call. If you don’t have a big field footprint or a lot of volunteers knocking on doors and making phone calls, the utility of coming up with really granular ways to break down the electorate just isn’t there. So the Romney campaign during the primaries was only present in a few states at a time. They were very savvy about early vote and absentee ballot handling, and they gamed those processes well. But they never made a robust investment in statistical modeling as Obama did in the 2008 primary. When you have volunteers in the field, they’re basically acting as your data collectors. And so you benefit from every time they talk to a voter. You’re not only learning about that voter, but you’re using those three or four data points from that interaction to inform your modeling about everybody else in the electorate. And because the Romney campaign did so little of that in the primaries, they never really built as ambitious and dynamic a cycle of modeling as the Obama campaign had in 2008. And since then they’ve been trying to ramp up nationally.
One great point of contrast in the ambition and scope of what they’re doing: In 2008 Barack Obama refreshed the electorate every week, nationally, which means every week, they would recalculate the probability of every one of the 170 million people they had in their records, and recalculate the probability of turning out to vote, of voting for Obama, and their views on certain issues. Romney in most states will probably only refresh those predictions once this whole general election.
So he’s that far behind even Obama’s 2008 machine?
Yes. Part of it’s the ambition. If you looked at the polling, you could say, ‘Look, people aren’t moving that much It’s a waste to go out in the field with 5,000 or 10,000 phone calls this week in New Mexico to recalculate individual-level probabilities.’ So there are good reasons to be skeptical of it as an investment. But it’s a reflection of how much the Obama campaign has invested in this. The fact is that if there is a dramatic shift in voter attitudes or likely behavior after one of these debates, the Obama micro-targeting model will almost certainly pick it up faster than the Romney campaign, and it will allow the ground-level decision-making in the Obama campaign to adjust more quickly in figuring out where their next round of mail will go, and which doors they’ll knock on and who they’ll call. And the big thing the Obama campaign did in 2008, they brought almost all of their capacity in-house instead of only relying on consultants and vendors. The Romney campaign has a few people coordinating things in-house, but they’re still generally relying on outside expertise and manpower. There are not enough people on the right who do this at a high level now to match what the left is capable is of.
What are the frontiers of the applications of this for the Obama campaign?
Persuasion experiments. So most of the experiments up until 2008 were fundamentally behavioral, focused on looking at changes in turnout or registration rates. The new frontier, and the Obama people are doing this, and they’ve contracted with the Analyst Institute to do the experimental design, is in what they call “experiment informed programs,” which is building randomized assignments of mail or web ads or field contact into their normal program. And then, because they’re polling continuously in large numbers, they can see who actually moved based on contact in the field. It basically frees you from relying solely on polls and focus groups to test messages. So instead of the traditional methods — you run a poll and ask people, “Would you be more or less likely to vote for Mitt Romney if you knew that companies he’d invested in at Bain had outsourced jobs?” — now you can go and randomly assign three different pieces of mail to three different groups — one that’s about Bain and outsourcing, and one that’s about Romney’s taxes, and one that’s about something altogether different — and then you’re polling across the population. And if people move more in response to one than the other, then you can attribute it to them having not just seen the message but having processed it and let it change their opinions. This is something the Obama campaign has been doing for a year, and the results are really remarkable. You start to realize how inadequate most of the usual tools for testing messages in politics are and how artificial focus groups or those type of poll questions are, because they’re relying on voters to predict what they might do if they got a piece of information in the future. This way, you can just give them the information and see what actually happens.
How does this compare to the state of innovation in this field in commercial marketing?
It’s a little different because it’s a different endgame. One thing that’s shifted in the last few years is, clearly a decade ago, people in politics looked to the corporate world and tried to figure out what lessons they could bring over from commercial marketing to politics. Now more and more you see people in the corporate world looking at statistical modeling and analytics in campaigns and hoping to take lessons from that.
As best I understand it, the credit card industry is still at the cutting edge of this, and in many ways, they’re risk modeling is comparable to politics, because there are downsides to missed targeting. For a lot of consumer marketers, there’s not a lot of downside to missed targeting. If I send a Coca Cola (KO) promotion to somebody who likes Pepsi (PEP), I’ve wasted however many dollars with that particular marketing contact, but I haven’t really lost anything. A credit card company knows that if they send a pre-approval notice to somebody who’s going to run up a $10,000 bill and not pay it off, they’ve made a grievous mistake. And the Obama campaign knows that if they send a piece of mail about his views on abortion to somebody who’s pro-life, or they have a volunteer knock on a door with a get-out-the-vote reminder the night before the election of somebody who’s a Romney voter, they’ve made a grievous mistake.
So in many ways predicting opinions in politics is a more delicate process than it is in the corporate world. It’s actually more refined. The Obama campaign in 2008 updated individual-level predictions weekly, and they’ve worked to try to make that even faster and more fluid and dynamic. I don’t think anybody at Coke or Subaru are really worried about adjusting their message to changes in consumer attitudes that way. And there’s just an idea of nimbleness in the way campaign activity works, where campaigns want to be more responsive to very small shifts in voter attitudes or behavior than most commercial marketers probably need to be.
Have you seen any evidence from the Obama White House that the political team integrated any of the lessons from the campaign into advancing their legislative agenda in the first term?
I don’t know. A lot of it has become best practices for folks who are organizing on the left, but most of the behaviors that are measured are around voting. There’s probably no reason why it couldn’t be translated for seeing if people call their Members. But almost all of the investment in research and testing has been around increasing rates of registering and voting. And so I think there’s far more science around campaigns than there is around legislative advocacy.
It seems like it’d be a ripe opportunity for interests groups on either side, like the AFL-CIO or the Chamber of Commerce.
I think that’s totally right. I’m just not sure anybody’s done it. One thing you can do — and I know people mess around with it, but I don’t know if it’s developed a lot of best practices — people are using patch-through calls to members of Congress. Part of what you need to do ultimately is track the behavior of your subject. Traditionally, if you told somebody to call their representative, you wouldn’t know who actually called. So it’s very hard to set up a randomized experiment that would measure that. And it’s very hard to measure the output of a statistical model if you can’t track whether people are actually doing things on the other end.
People now use these patch-through calls to Congress, and that’s a place where you can micro-target people off of a voter file, and you can call them and say, “Please hold, I’ll connect you to Sen. Inhofe,” and you can actually know how long they stayed on the line, or whatever else. So it’s certainly testable and there are trade associations and lobbying interests in Washington who are trying to do more with micro-targeting analysis.
Are you walking away from writing this book more or less hopeful about our politics and our democracy? You write in an optimistic way about how aspects of these innovations have encouraged people to return to a 19th-century model of face-to-face voter contact and engagement. But the science of it, while proven, also seems kind of cynical. What the campaigns are doing to manipulate people — there’s a stark gap between that and the rhetoric you hear from candidates about how they view the American people.
You read that optimism correctly. Part of it is politics was worse a decade ago. Candidates and parties would write off whole cities or counties because they didn’t know how to pick out the 30% of people who were favorable to their message. So those people just by geography were counted out of the equation that the campaigns looked at when they were trying to figure out how to win. So one thing, just on the data and micro-targeting side that’s unquestionably positive is that campaigns now feel that they can put together a winning coalition across traditional demographic boundaries and across geographic lines because they can assess individual likelihoods of being responsive.
We talk a lot about how demographics have made states like Virginia and North Carolina newly competitive for Democrats, but you need the tools to put together those coalitions. That’s possible because campaigns on both sides now engage individuals on their own terms and don’t have to write them off depending on where they live. That I think is indisputably good.
On the behavioral science stuff, even the things that look cynical have been in service of getting people to vote. So it’s hard for me to get really down about that. We don’t have compulsory voting in this country, and there are lots of good reasons people have for not wanting to vote. So if campaigns are using psychological tricks to remind people of the good reasons to vote, that doesn’t seem like a bad thing. I’d be worried if people were using behavioral psychology, insights and methods to demotivate voters. And there’s frankly no reason why you couldn’t turn a lot of these techniques on their head and use them to convince people that they shouldn’t vote. But that shouldn’t be the case, and it’s not what’s happened.
So I think when you combine those two things, you have a lot more data that allows campaigns to feel they understand a voter’s attitude. The other thing is campaigns use micro-targeting to figure out who they should talk to and how they can most meaningfully engage them. And so I think you tend to have a more meaningful, substantive interaction with voters if you know who they are. And if the choice is between having campaign contact that’s directed by a sophisticated modern understanding of the human brain as opposed to a more primitive one, that seems all good to me.
A shorter version of this story appeared in the October 29, 2012 issue of Fortune.