Microtargeting Myth vs. Fact
BY Patrick Ruffini | Friday, July 18 2008
There's a lot of hype surrounding microtargeting -- which is the process of targeting voters scientifically based on consumer and demographic data. This piece in Salon yesterday on "Obama's super marketing machine" is no different. But as someone with a bit more than a passing understanding of what microtargeting is, I have to shake my head a little at articles like this. Because the media gets it almost completely wrong -- whether it's hyping relatively mundane technologies or celebrating "sexy" examples (dial up 40 year old Vodka drinking Volvo drivers!) that have almost no bearing on microtargeting's usefulness in real life.
The Salon story is a textbook "what George Bush/Barack Obama/John McCain knows about you" approach to the topic, of the kind we've seen on a pretty regular basis since 2004. Let's start the nitpick from the beginning:
About every week or so, you get an e-mail from Barack Obama campaign manager David Plouffe, or top deputy Steve Hildebrand, or maybe Obama himself. They're breezy and informal, addressing you by first name at the outset (before they ask you to donate money at the end). But that's just the beginning.
You know, of course, that Obama has your e-mail address. You may not have realized that he probably also has your phone number and knows where you're registered to vote -- including whether that's a house or an apartment building, and whether you rent or own. He's got a decent estimate of your household income and whether you opened a credit card recently. He knows how many kids you're likely to have and what you do for a living. He knows what magazines and catalogs you get and whether you're more apt to get your news from cable TV, the local newspaper or online. And he knows what time of day you tend to get around to plowing through your in box and responding to messages.
The 5 million people on Obama's e-mail list are just the start of what political strategists say is one of the most sophisticated voter databases ever built.
There is nothing technically wrong with these claims, but conflating Obama's 5 million person email list (thanks to Mike Madden for confirming its size) with microtargeting is a bit of a stretch. To get on Obama's email list, all you have to give is your e-mail and ZIP code. Campaigns keep this registration process fairly simple to attract the most possible signups. But as a result, it is nearly impossible to match these records to consumer data without something more concrete, like a first & last name, and hopefully a street address.
Is Obama sending you emails tailored to your individual tastes, beyond simple registration or demographic preferences? Unlikely. E-mail is the cheapest form of communication, and so it's not necessarily cost-effective to microtarget. I also have not encountered an Obama national email circulated on the Internet that is different than the versions I have personally received. You microtarget on stuff that's expensive, like phones.
Madden goes on to claim that Obama's operation is far more sophisticated than Bush-Cheney 2004:
The sheer scale of the operation -- because of Obama's large network of supporters and heavy emphasis on field organizing -- means the data can be sliced in ways that the Bush-Cheney campaign couldn't have dreamed of in 2004. It's most likely also more advanced than what either side did in the 2006 elections, or, for that matter, what John McCain is doing now.
Is this Obama's campaign vastly more capable in the microtargeting realm than Bush's was? I can't tell from this article. That's because the processes it describes were either around in 2004 or are relatively rudimentary Web technologies embedded in most technology platforms, including Drupal, which runs TechPresident.
This gives us a basic primer on what microtargeting is:
All that data gathered in one place may seem a little spooky, though the average credit card company already has it and then some. Ultimately, the approach is about greater sophistication and efficiency. It means the campaign may not wind up wasting time contacting people who are probably voting for McCain, and that when Obama aides or volunteers go out looking for supporters, they have a pretty good idea of what issues those potential supporters care about most. It's the political equivalent of what big corporate marketers have been doing for years: If you're a baby boomer living in Westchester County, N.Y., golf gear catalogs will show up in your mailbox, but if you're a 20-something living in Williamsburg, Brooklyn, you might get a free trial of Spin magazine instead. Now the same goes for politics -- if you're in a demographic that makes you statistically likely to have children, Obama might send you an e-mail about education policy instead of one about taxes.
No argument here. Just thought it would be helpful to have in for your benefit.
What goes into these microtargeting files?
if you've ever registered a product -- a TV, a computer or a microwave, for example -- chances are the campaign knows you own it. Likewise, they know if you've signed up for the frequent customer club at your local Whole Foods, or if you've joined the American Civil Liberties Union. (Yes, those last two probably make you an Obama supporter). Or whether you own a gun and have a current hunting license. (An indicator you're less likely to pull the lever for him in November.)
But as a Democratic strategist (wisely) explains none of these variables alone tell us much. Yes, the campaign may "know" you own a particular model TV -- actually, something that granular may be abstracted into some broader, more useful variable given the multitude of TVs out there -- but it's not like they're using it to spy on you. It's just a blip in a dataset somewhere that no one will likely every physically set eyes on.
"People get hung up looking for a silver bullet," said Ken Strasma, a Democratic consultant whose firm, Strategic Telemetry, worked on more than 100 races in 2006 and is mining data for Obama now. "They want to know, is it cat owners or bourbon drinkers or some nice buzz phrase like that. It's when you see the interactions between hundreds of different data points [that patterns emerge] -- it's rare that you see one single indicator pop."
Now, it gets interesting. Madden starts getting into the meat of how Obama may have built a better mousetrap:
Now Obama's campaign is aiming to be ahead of even the GOP's standard in applying sophisticated data mining techniques across the board, supported by all the traditional canvassing, door-knocking and other work it's been doing. The campaign is collecting some of the most helpful data on its own. For example, aides can track what time you open e-mails from them, and if you show a consistent pattern, they'll start sending them at around that time of day. "The marginal benefit of sending some people an email at 2 o'clock vs. 3 o'clock vs. 4 o'clock might not make sense [at first]," said Michael Bassik, a Democratic consultant with MSHC Partners, the firm that did John Kerry's online advertising in 2004. "But once you start getting an e-mail list that's 3 million, 4 million, or 10 million people, increasing the returns for a fundraising e-mail by 5 or 10 percent means additional returns of tens or hundreds of thousands of dollars."
There are no specifics on how Obama is using door knocks to enhance microtargeting, so this claim is difficult to evaluate (I can tell you this concept is not exactly unknown in Republican circles, though). But the bit about timing emails is interesting.
Could Obama be timing his emails to different segments of his list depending on when they open? It's definitely an interesting possibility, and definitely doable (though a lot of work). The other day, Soren forwarded me an email he had received from Obama at 6 in the morning. I received the same email at around 8. Being more of a night owl, I guess that could explain it -- but it's also true that sending to a 5 million person list is a massive task that takes several hours, so the discrepancy isn't prima facie evidence of anything. If the send-timing thing is true, it's very smart. At a minimum, campaigns should send at times of day when supporters are statistically more likely to open (usually 9am), and send first to the people most likely to take action.
This starts getting into the mundane:
If you're one of the 1 million people who have a login on Obama's social networking site, they know how often and when you visit, and they can use that to gauge how committed you are to the campaign. A few months ago, the campaign sent out a three-page survey asking people about their voting habits, how often they go to church, which groups and issues they identify with and whether they've given money to political candidates in the past. The point of all of the online gadgetry is to get people to show up for offline events. "We've tried to orient the tools less as a social network and more as a mobilization network," said Joe Rospars, Obama's online director. "We're creating opportunities for people to get out there and do things -- the campaign is election-outcome oriented."
Wow! A website that tracks when you've logged in, how long you've been, and how often you come back. Psst, but the rest of us techies call that an access log! There is nothing remarkable about this technology. What is rarer is people using it effectively. My dashboard on My.BarackObama.com tells me my rank based on my (limited) activity on the site, and having logged in a few times, blogged once, set up a fundraising page for demonstration purposes, and joined a couple of groups, I am ranked 88,432nd, or in the top 10%. This is just a matter of reading rows from a database and assigning a point value for each one. You can see my access log below:
I've also analyzed how My.BarackObama.com is driving people to take action offline. But the point is that this is hardly new and innovative. Both Bush and Kerry had systems like this in '04, and this type of activity streaming comes standard with most social networking platforms nowadays.
Finally, and this is credit-worthy, is this bit:
Offline, volunteers are canvassing neighborhoods where they think they'll find supporters, or getting contact info at Obama's big rallies, picking up chunks of similar data. Unlike with previous campaigns, Obama's aides dump all the information they get into one centralized database. So if you give the campaign $50 from an online solicitation, then show up at a rally organized offline, the campaign knows that. Likewise, if you join Obama's Facebook group (approximately 1 million strong), then later buy an Obama '08 umbrella, aides file that away for possible use later.
The part about centralized databases is big, and not to be underestimated. But without being able to peek under the hood, it's hard to tell whether this is more or less sophisticated than what, say, the RNC has -- as they've been dealing with the same issues at a very high level of complexity. Getting different databases to sync up properly is a big challenge, and if Obama has figured out a fix, hats off to them.
The part about Facebook is likely a generalization. It is impossible to harvest Facebook fan or group data and the existing technologies don't let you get into the nitty-gritty of who joined what groups, or more aptly, who signed up on Obama's page. It is impossible to vector this data (beyond simply knowing that someone is signed up on Facebook) with in-house data, which is one of the frustrations of using Facebook for political organizing -- they own the data, not the campaign.
A broader point to be made about articles like this is that they emphasize the whiz-bang aspects of microtargeting: here's a segment of Whole Foods-loving Obama umbrella owners! But those narrow segments are rarely of use to a real-world campaign. Used properly, microtargeting is all about mining of data points to categorize people into broad buckets (R, D, I) or (social / economic / national security conservative) when the existing methods fail (hard ID, survey responses, party registration, etc.)
As I've written before, there is a danger in over-segmenting the electorate. And this is where microtargeting goes wrong. If you're using the technology to divine minute differences between 80 different segments of the electorate and then sending them 80 different messages, you need your head examined. This is Mark Penn's microtrends gone amuck. Not only are you wasting time with 20 or 30 times amount of work you need to be doing, but if you try to communicate everything, you end up communicating nothing. As brilliant as the Obama/Bush microtargeting model is, both candidates understood the power of central, unifying messages that cut through the clutter. Why is the word most associated with Obama "change"? Message discipline!
A better model for microtargeting is to use it to find your One True Supporter. Say that we know that the ideal Obama donor is a Mac-addled, latte-sipping urbanite. Then throw all your resources at finding more of those people (a successful application of this would have been both parties trolling the iPhone lines for new tech talent), rather than persuading reluctant middle aged moms to join the ranks. Done right, microtargeting can be used to reinforce your one true brand, rather than splinter it.