How Far Can Political Technology Reach?

Digital tools are becoming more and more central to the process of winning your vote. Just how far can technology reach?

Can only be used for the article by Andrew Rice that ran in the 7/19/2014 Natinoal Journal magazine. 
National Journal
Add to Briefcase
Andrew Rice
July 18, 2014, 1 a.m.

If you are a liv­ing, breath­ing, vot­ing-age Amer­ic­an, the polit­ic­al ana­lyt­ics firm i360 says it has your num­ber. Its data­base con­tains more than 250 mil­lion in­di­vidu­al pro­files, ba­sic­ally one for every adult in the coun­try, each of which is scored on a 100-point scale. Zero rep­res­ents the most par­tis­an Demo­crat, and 100 the most par­tis­an Re­pub­lic­an. The high­er the score, the bet­ter it is for i360, which is af­fil­i­ated with the con­ser­vat­ive non­profit net­work fun­ded by the Koch broth­ers. The firm claims that its pre­dict­ive mod­el helps its cli­ents to judge wheth­er any giv­en per­son will likely be re­cept­ive to an ad­vert­ise­ment or a knock at the door — and will ul­ti­mately be in­clined to vote the right way.

The com­pany pin­points a per­son’s place on its spec­trum by us­ing in­form­a­tion from sources ob­vi­ous and ar­cane: voter re­gis­tries, cred­it agen­cies, so­cial net­works, and so on. Are you rich or poor? Do you live in a heav­ily Demo­crat­ic pre­cinct? Are you re­lated to a lot of Re­pub­lic­ans? Have you tweeted dis­pleas­ure with Pres­id­ent Obama? Did you re­cently pur­chase a Bible? Any of these data points might move the needle, i360 says.

Like most play­ers in the world of di­git­al polit­ics, i360 main­tains an air of mys­tery around the in­ner mech­an­ics of its al­gorithms, and the firm keeps a low pro­file, rarely grant­ing in­ter­views. It re­cently moved in­to a new of­fice in sub­urb­an Wash­ing­ton, which I was asked not to de­scribe in any fur­ther de­tail. Al­though its pres­id­ent, Mi­chael Palmer, de­clined to dis­cuss spe­cif­ic cli­ents, pre­vi­ous re­ports and pub­lic re­cords in­dic­ate a tight re­la­tion­ship with the Koch-sup­por­ted or­gan­iz­a­tion Free­dom Part­ners and an af­fil­i­ated non­profit called the Themis Trust, which has spent mil­lions to build an in­de­pend­ent ana­lyt­ic­al ca­pa­city. “An out­side en­tity can in­vest in the right things, with a long-term vis­ion,” Palmer says. “A lot of the stuff that we’re build­ing right now isn’t go­ing to be ready un­til 2016.”

Of course, the ba­sic idea be­hind i360 — the no­tion that tech­no­logy can open a portal in­to voters’ minds — is noth­ing new. Polit­ic­al pro­fes­sion­als, who work at the nex­us of ideo­logy and mar­ket­ing, have been de­vel­op­ing mi­crotar­get­ing tech­niques for years. And in 2012, Obama cam­paign staffers — work­ing from a labor­at­ory called “the Cave” — fam­ously mined data to new ex­tents of gran­u­lar pre­ci­sion.

(Justin Sullivan/Getty Images) National Journal

What has changed since then — and con­tin­ues to change, al­most by the day — is the sheer num­ber of firms and con­sult­ants op­er­at­ing in the polit­ic­al-tech­no­logy sec­tor. Huge sums of cash are flow­ing in­to it. Parties are fun­drais­ing for it. Bil­lion­aire ven­ture cap­it­al­ists are in­vest­ing in it. Start-ups that spe­cial­ize in it are con­stantly be­ing born. With one eye on the midterms and the oth­er on 2016, polit­ic­al tech­no­lo­gists are con­struct­ing new weapons and ex­per­i­ment­ing with in­creas­ingly per­son­al­ized tar­get­ing. And the com­pet­i­tion between Demo­crats and Re­pub­lic­ans is sure to in­tensi­fy as they head in­to a wide-open pres­id­en­tial race, where bil­lions of dol­lars will be spent un­der rad­ic­ally loosened cam­paign fin­ance reg­u­la­tions.

Re­pub­lic­ans, in par­tic­u­lar, have been ag­gress­ively press­ing to re­verse-en­gin­eer the Obama cam­paign’s tac­tics. “They it­er­ated for­ward, but I don’t think the vis­ion has been fully real­ized yet,” Chuck De­Feo, the Re­pub­lic­an Na­tion­al Com­mit­tee’s chief di­git­al of­ficer, said in a present­a­tion at a re­cent tech­no­logy con­fer­ence called the Per­son­al Demo­cracy For­um. He de­scribed an R&D ef­fort that in­cludes a Sil­ic­on Val­ley “in­cub­at­or,” run by an en­gin­eer re­cruited from Face­book, that is re­portedly at work on a com­puter sys­tem called Beacon. But as is of­ten the case with ad­vanced tech­no­logy, the spe­cif­ics re­main dif­fi­cult for out­siders to eval­u­ate. “For the first time, I’m go­ing to pub­licly show what that plat­form looks like,” De­Feo prom­ised the audi­ence, but he was just jok­ing. When he flipped to his next slide, the gi­ant screen be­hind him pro­jec­ted the men­acing im­age of the Death Star.

For all the fu­tur­ist­ic rhet­or­ic that sur­rounds the grow­ing role of tech­no­logy in polit­ics, the ac­tu­al cap­ab­il­it­ies of the new tools aren’t well un­der­stood — not by the gen­er­al pub­lic, and not even by many people who work in polit­ics. “There’s a lot of snake-oil sales­men com­ing through, tak­ing ad­vant­age of the lack of know­ledge,” says Zac Mof­fatt, the di­git­al dir­ect­or for Mitt Rom­ney’s 2012 cam­paign who now runs the firm Tar­geted Vic­tory. Polit­ics, ad­vert­ising, and the tech busi­ness are all prone to ex­ag­ger­a­tion, and when they come to­geth­er, the huck­ster­ism can get gran­di­ose. Poke around the sub­ject, and your email in-box will soon be­gin to fill with sub­ject lines that read like ads for male en­hance­ment: “Deep Root Ana­lyt­ics: 5 ex­tra points at the polls.”

Can any tech­no­logy really de­liv­er those kinds of res­ults? And if so, how does it ac­tu­ally work? The di­git­al sphere of polit­ics en­com­passes a wide vari­ety of func­tions, in­clud­ing data-gath­er­ing, ana­lys­is, and ad tech­no­logy — and some of them are bet­ter de­veloped than oth­ers. There’s no doubt that there’s been an ex­plo­sion in the amount of data avail­able to cam­paigns, as every vari­ety of hu­man in­ter­ac­tion — con­ver­sa­tion, com­merce, friend­ship, en­ter­tain­ment — moves in­to the in­cess­antly mon­itored and quan­ti­fied on­line realm. And it’s equally clear that the Obama cam­paign’s in­nov­a­tions will leave a leg­acy as the vol­ley that set off an arms race. But as po­tent as their sur­veil­lance may be, the tech­no­lo­gists still face a chal­lenge: How well can they trans­late everything they know about voters in­to votes?

“There’s a lot of snake-oil sales­men com­ing through, tak­ing ad­vant­age of the lack of know­ledge.”

THE IN­NOV­AT­ORS ARE al­ways work­ing around a cent­ral irony: The very ad­vances that make it pos­sible to know so much about voters also make them more dif­fi­cult to reach. A DVR re­cords your view­ing habits, but it also al­lows you to fast-for­ward through the stand­ard 30-second cam­paign spot. Spam fil­ters are rising; net­work audi­ence num­bers are fall­ing. It takes plenty of in­ven­tion just to coun­ter­act the re­lent­less force of me­dia en­tropy.

The pro­cess of find­ing and per­suad­ing a po­ten­tial voter pro­gresses in stages; con­sumer mar­keters, who use sim­il­ar meth­ods, em­ploy the meta­phor of the “pur­chase fun­nel.” At the top of the fun­nel sit firms that gath­er and ana­lyze data, like the Demo­crat­ic power­houses Ca­tal­ist and NGP VAN, or i360, which is try­ing to re­pro­duce the mod­el for Re­pub­lic­ans. The ana­lyt­ics firms, in ef­fect, de­scribe voters to cam­paigns, help­ing them to size up the mar­ket­place for their mes­sages. Al­though the meth­ods they em­ploy are alarm­ing to some pri­vacy ad­voc­ates, they are per­fectly leg­al, and both parties use them avidly.

Like Google, the Na­tion­al Se­cur­ity Agency, or the Demo­crat­ic data ma­chine, i360 has a vo­ra­cious ap­pet­ite for per­son­al in­form­a­tion. It is con­stantly in­gest­ing new data in­to its tar­get­ing sys­tems, which pre­dict not only par­tis­an iden­ti­fic­a­tion but also sen­ti­ments about is­sues such as abor­tion, taxes, and health care. When I vis­ited the i360 of­fice, an em­ploy­ee gave me a demon­stra­tion, zoom­ing in on a map to fo­cus on a par­tic­u­lar 66-year-old high school teach­er who lives in an apart­ment com­plex in Al­ex­an­dria, Vir­gin­ia. If a cam­paign wanted to reach her, it could send a vo­lun­teer to re­cord her opin­ions on i360’s smart­phone app, while track­ing the en­counter via GPS. If she seemed re­cept­ive, the cam­paign could use what con­sumer mar­keters call “look-alike mod­el­ing” to loc­ate oth­ers with sim­il­ar pro­files. Though the ad­vert­ising in­dustry typ­ic­ally es­chews ad­dress­ing any single in­di­vidu­al — it’s not just in­vas­ive, it’s also in­ef­fi­cient — it is be­com­ing com­mon­place to tar­get ex­tremely nar­row audi­ences. So the school­teach­er, along with a few look-alikes, might see a tailored ad the next time she clicks on You­Tube.

“We say every dif­fer­ent demo­graph­ic and con­stitu­ency can be reached through dif­fer­ent means, at dif­fer­ent times in the elec­tion, but here’s the people you need to talk to,” Palmer says. “Is this a per­fect sci­ence? Do I know ex­actly what is go­ing to move you? No, but I have a high­er de­gree of prob­ab­il­ity of mov­ing you us­ing this data.”

{{third­PartyEmbed type:magazineAd source:magazine_mid}}

The in­flu­ence of ana­lyt­ics has ris­en, in part, to com­pensate for the in­creas­ingly glar­ing fail­ings of tra­di­tion­al tele­phone polls. About a third of house­holds use only cell phones, mak­ing them tough to call, and poll re­sponse rates on land­lines have fallen from about 40 per­cent to less than 10 per­cent since the 1990s. Small samples re­quire much stat­ist­ic­al tweak­ing to yield ac­cur­ate res­ults. “Every single cycle it gets more and more dif­fi­cult,” says Dav­id Roth­schild, an eco­nom­ist who has stud­ied polling for Mi­crosoft Re­search. “As re­sponse rates fall fur­ther and fur­ther, it be­comes more likely that they may miss that crit­ic­al point where the sample is no longer more or less rep­res­ent­at­ive. And once that hap­pens, it is not go­ing to work any­more.” 

The less rep­res­ent­at­ive the sample, the great­er room for er­ror — or even wish­ful think­ing — to creep in. Rom­ney’s team, to its sub­sequent hu­mi­li­ation, be­lieved in­tern­al polls that showed him run­ning ahead in sev­er­al battle­ground states on the eve of the 2012 elec­tion. The Obama cam­paign, by con­trast, at­temp­ted to fill the know­ledge gap with ana­lyt­ics. Many cap­ab­il­it­ies that firms like i360 now tout were de­veloped by the pres­id­ent’s team in 2012, which built on ad­vances the Demo­crats made in 2008, which in turn drew from the mi­crotar­get­ing suc­cess of George W. Bush in 2004. Fa­cing a tough reelec­tion chal­lenge, Obama’s cam­paign strove to un­der­stand the elect­or­ate “at an atom­ic level,” the journ­al­ist Sasha Is­sen­berg re­counts in his book The Vic­tory Lab. “Obama’s cam­paign began the elec­tion year con­fid­ent it knew the name of every one of the 69,456,897 Amer­ic­ans who had put him in the White House,” Is­sen­berg writes. “They would re­as­semble the co­ali­tion, one by one, through per­son­al con­tacts.”

“We say every dif­fer­ent demo­graph­ic and con­stitu­ency can be reached through dif­fer­ent means, at dif­fer­ent times in the elec­tion, but here’s the people you need to talk to.”

To carry out that mis­sion, Obama put to­geth­er a huge ground op­er­a­tion, backed by an ana­lyt­ics staff of more than 50 people — 10 times the size of Rom­ney’s — and a team of em­bed­ded so­cial sci­ent­ists that tested the per­suas­ive power of the cam­paign’s mes­sage. In­stead of just re­ly­ing on tele­phone polls of sup­posedly likely voters, it pro­duced re­ports that at­temp­ted to gauge the sen­ti­ments and demo­graph­ic makeup of the elect­or­ate based on hun­dreds of data points. “The theme is that there are more di­verse means of meas­ure­ment,” says Dan Wag­n­er, an Obama cam­paign num­ber-crunch­er who, with fund­ing from Google’s Eric Schmidt, sub­sequently foun­ded a start-up called Civis Ana­lyt­ics. “The thing about on­line is, you can col­lect data very cheaply, very widely, and very quickly.”

At least in the­ory, polit­ic­al cam­paigns start with a good idea of who might be buy­ing their product. Voter re­gis­tra­tions are pub­lic re­cord, and both parties main­tain vo­lu­min­ous files of names. But as a prac­tic­al mat­ter, re­gis­tra­tion in­form­a­tion is dis­persed haphaz­ardly across 50 states and must be con­stantly re­freshed. And even then, it can only re­veal so much about how someone is go­ing to vote.

That’s where ana­lyt­ics can help cam­paigns fig­ure out the elect­or­ate. Crude tech­niques have ex­is­ted for years: fil­ter­ing the voter file ac­cord­ing to demo­graph­ics such as age or race, or us­ing his­tor­ic­al pre­cinct res­ults, or cross-ref­er­en­cing the mem­ber­ship lists of or­gan­iz­a­tions like Planned Par­ent­hood and the NRA. At a more soph­ist­ic­ated level, cam­paigns can draw on reser­voirs of in­di­vidu­al con­sumer data that agen­cies such as Ex­peri­an use to com­pile cred­it scores. There are ways to bore in on spe­cif­ic types of pur­chases. But in­dic­at­ors such as in­come level and gun own­er­ship are still just prox­ies. It was only re­cently that ana­lysts began to be able to be tap in­to a much more re­li­able gauge of pub­lic opin­ion: what people re­veal about them­selves when they ima­gine they are just among friends.

(Jewel Samad/Getty Images) Jewel Samad / Getty

(Jew­el Sa­mad/Getty Im­ages)”The group of people on Face­book are not rep­res­ent­at­ive of the coun­try yet, but they are rep­res­ent­at­ive of all those people on Face­book,” Wag­n­er says. What makes a site like Face­book valu­able to ad­vert­isers is the in­form­a­tion it con­tinu­ally gath­ers, even when you don’t think it’s look­ing. “The thing about the new so­cial net­works is that they are cre­at­ing un­pre­ced­en­ted con­sumer pro­files,” says Max Kale­hoff, seni­or vice pres­id­ent at So­cial­Code, a com­mer­cial ad tech com­pany. “Most people are logged onto Face­book whenev­er they do any­thing else on the Web. It cre­ates this in­cred­ible bread-crumb trail.”

Many of the ba­sic tech­niques used by Obama’s ana­lysts were ori­gin­ally de­veloped in the private sec­tor — by ad tech com­pan­ies that si­lently re­cord and cat­egor­ize on­line be­ha­vi­or, re­du­cing in­di­vidu­al con­sumers to sets of data points. Ad tech has ad­vanced with great leaps in re­cent years, bring­ing the in­dustry close to its om­ni­scient goal: be­ing able to re­li­ably pin­point what a single per­son is up to on his or her com­puter, TV, and phone, all the time. Mo­bile tech­no­logy, in par­tic­u­lar, has opened many new mon­it­or­ing op­por­tun­it­ies. If you have a so­cial-net­work­ing app on your phone, it may know all sorts of things about your daily in­ter­ac­tions, wheth­er vir­tu­al or lit­er­al. “It con­nects every­body through those plat­forms pretty much 24 hours a day,” Wag­n­er says. “The device is al­most ir­rel­ev­ant. What’s valu­able is the net­work.” 

While the amount of per­son­al in­form­a­tion avail­able to cam­paigns keeps ex­pand­ing, however, the lim­its of the pub­lic’s tol­er­ance for its use are still be­ing tested. Face­book re­cently ran in­to con­tro­versy when it was re­vealed that its Data Sci­ence di­vi­sion was ex­per­i­ment­ing with in­stilling pos­it­ive and neg­at­ive moods in users through their news feeds. Politi­cians are un­der­stand­ably wary of look­ing like spies. A 2012 Uni­versity of Pennsylvania study found that 64 per­cent of Amer­ic­ans would be less likely to vote for a can­did­ate if they knew he or she was buy­ing data about their on­line be­ha­vi­or.

{{third­PartyEmbed type:magazineAd source:magazine_bot­tom}}

Yet even as people com­plain about in­tru­sion, any user of Face­book or Twit­ter knows there is little lim­it to what many will share about their lives and opin­ions. Roth­schild, the Mi­crosoft re­search­er, has been ex­per­i­ment­ing with ways to put this con­nectiv­ity to use, through polling sys­tems that rely on re­sponses to sur­veys on devices like the Xbox. In a forth­com­ing pa­per, au­thored in co­oper­a­tion with schol­ars at Columbia Uni­versity, Roth­schild writes that even an un­rep­res­ent­at­ive pop­u­la­tion such as video-game play­ers could be made to yield stat­ist­ic­ally re­li­able res­ults if the sample size was large enough. “I think you’ll see an ex­plo­sion of this type of polling in 2016,” says the eco­nom­ist, who has dis­cussed his work with op­er­at­ives on both sides of the aisle.

However, Roth­schild says, there’s a prac­tic­al ques­tion that re­mains un­answered: What’s the value of all this in­form­a­tion? “We’re do­ing a good job at col­lect­ing data and a de­cent job at ana­lyz­ing it,” he ar­gues. “But we’re do­ing a really me­diocre job at us­ing that data.”

ONCE A CAM­PAIGN has com­piled its list of tar­gets, there are still only a few ways to de­liv­er the polit­ic­al pay­load, and none of them are really new. The first is through a mail­box, either postal or di­git­al. Polit­ic­al con­sult­ants have built a luc­rat­ive in­dustry around dir­ect mail, but its mod­ern im­pact is de­bat­able. Rom­ney over­whelm­ingly out­spent Obama on dir­ect mail. Mean­while, Bill de Bla­sio won last year’s com­pet­it­ive New York City may­or­al primary without send­ing out a single mass mail­ing.

The second way to de­liv­er mes­sages is through per­son­al in­ter­ac­tions. Voters can be con­vinced by vo­lun­teers at their doors or friends in the gro­cery aisle — or on Face­book. Tra­di­tion­ally, door-to-door can­vassing has been blue-col­lar work. But tech­no­logy has upen­ded that hier­archy by giv­ing a fu­tur­ist­ic sheen to grass­roots mo­bil­iz­a­tion. So­cial net­works make it pos­sible to co­ordin­ate per­son­al in­ter­ac­tions on a massive scale. Jim Gil­li­am, founder of Na­tion­Build­er, a non­par­tis­an or­gan­iz­ing plat­form, says his tech­no­logy al­lows cam­paigns to fol­low what people are say­ing on­line, so they can ul­ti­mately en­list core sup­port­ers to re­cruit their friends. “It all flows in­to one data­base,” he says. “Ba­sic­ally, what’s go­ing to hap­pen is that any­body who ex­presses in­terest in the most mod­est way can be fol­lowed up with by someone who cares.”

“TV is ripe for dis­rup­tion. There are so many in­ef­fi­cien­cies to ex­ploit.”

Na­tion­Build­er is fun­ded by mil­lions of dol­lars in ven­ture cap­it­al from tech in­vestors such as Marc An­dreessen and Pierre Om­idy­ar. Gil­li­am has a back­ground in pro­gress­ive causes, but the com­pany does busi­ness with any cli­ent that is will­ing to pay, which makes Na­tion­Build­er both a rar­ity in polit­ic­al circles and an ob­ject of some sus­pi­cion. (The com­pany’s ag­nosti­cism has also made it a tar­get for hack­ers, who re­cently brought down its net­work with what Gil­li­am de­scribed as an at­tack “dir­ec­ted at one of our cus­tom­ers for their polit­ic­al be­liefs” and “meant to dis­rupt up­com­ing elec­tions.” I was told the cus­tom­er was Bri­tain’s anti-im­mig­rant party UKIP.) Gil­li­am re­jects the no­tion that pro­gram­ming should be tied up in par­tis­an polit­ics. “It’s just not pos­sible to get that tech­no­logy sol­id in the high-pres­sure en­vir­on­ment of a cam­paign, where you ba­sic­ally don’t care if it works after Elec­tion Day,” he says. “That’s why we’re struc­tured as a tech com­pany.”

In­deed, polit­ic­al cam­paigns form and dis­band quickly, and in­fight­ing is rife in­side party com­mit­tees, with mil­lions in cam­paign spend­ing at stake. Re­cently, The Huff­ing­ton Post re­por­ted that Andy Bar­kett, the chief tech­no­logy of­ficer the RNC hired away from Face­book with much fan­fare, had already been side­lined amid dis­sen­sion. Even the vaunted Obama cam­paign had its share of in­tern­al turf battles — and struggled at times to trans­late the ana­lyt­ics per­formed in the Cave in­to on-the-ground ac­tion. “It’s a little bit the un­told story of the Obama cam­paign,” Gil­li­am says, “be­cause it’s just not as sexy as, ‘There was this gi­gant­ic com­puter data­base that was col­lect­ing votes.’ “

One of the Obama cam­paign’s most-hyped in­nov­a­tions was its use of “tar­geted shar­ing” — ba­sic­ally, a form of friend-to-friend mo­bil­iz­a­tion ginned up through a cus­tom­ized Face­book app. The cam­paign boas­ted that it had gen­er­ated 5 mil­lion con­tacts through tar­geted shar­ing in 2012, and af­ter­ward it be­came a pop­u­lar buzzword in con­sult­ing circles. But Mof­fatt, the former Rom­ney strategist, points out that a pair of aca­dem­ics later pro­duced an es­tim­ate that the pro­gram may have res­ul­ted in a mere 20,000 votes for the pres­id­ent. “Postelec­tion, there’s al­ways the shiny ob­ject,” Mof­fatt says. “Con­cep­tu­ally it sounds so cool that people are will­ing to pay as­tro­nom­ic­al amounts for it.”

Andy Barkett speaks during the Republican Jewish Coalition spring leadership meeting. (Ethan Miller/Getty Images) Ethan Miller / Getty

Andy Bar­kett speaks dur­ing the Re­pub­lic­an Jew­ish Co­ali­tion spring lead­er­ship meet­ing. (Eth­an Miller/Getty Im­ages)US­ING SO­CIAL NET­WORKS ;to muster vo­lun­teer armies sounds great at con­fer­ences, but, in prac­tice, polit­ic­al op­er­at­ives know you tend to get what you pay for. Ul­ti­mately, if a cam­paign wants to make its case, it needs to ad­vert­ise. For most large cam­paigns, the paid me­dia budget is the largest area of ex­pendit­ure, and typ­ic­ally around 90 per­cent of it is spent on TV. But an un­told pro­por­tion of that money is wasted. A re­cent sur­vey con­duc­ted for Google and sev­er­al polit­ic­al con­sult­ing firms found that over the course of the pri­or week, one-third of likely voters had watched no tele­vi­sion pro­gram­ming at the time it was broad­cast, out­side of sports.

That still leaves two-thirds of the elect­or­ate that can be reached through tele­vi­sion, but they are now spread across hun­dreds of avail­able chan­nels, which charge ad­vert­isers widely vary­ing rates for air­time. In 2012, the Obama cam­paign tried to fig­ure out how to spend its tele­vi­sion budget more ef­fi­ciently, de­vel­op­ing something called the Op­tim­izer. This sys­tem cross-ref­er­enced the names of cam­paign tar­gets with data com­piled by a com­pany named Ren­trak, which has ac­cess to mil­lions of cable boxes. In­stead of simply buy­ing ads on CNN and the nightly news, like most can­did­ates do, the Obama cam­paign figured out what shows its de­sired audi­ence was watch­ing, and how to reach them at the most cost-ef­fect­ive time slots.

“The Rom­ney cam­paign out­spent us 8-to-1 on the news be­cause they were work­ing from the old play­book,” says Chaun­cey McLean, who worked on the Obama cam­paign and, after the elec­tion, cofoun­ded a com­pany called the Ana­lyt­ics Me­dia Group to re­fine the Op­tim­izer’s tech­no­logy and mar­ket it to polit­ic­al cam­paigns and com­mer­cial brands. When I vis­ited McLean’s New York City of­fice re­cently, he showed me a re­port pre­pared for a can­did­ate in the San Diego tele­vi­sion mar­ket, which re­com­men­ded buy­ing late-night re­runs of shows like Lever­age and Fam­ily Guy. Since 2012, the tac­tic has be­come com­mon­place, and Re­pub­lic­an firms such as i360 have formed sim­il­ar part­ner­ships with Ren­trak.

“TV is ripe for dis­rup­tion,” says McLean’s part­ner Chris From­mann. “There are so many in­ef­fi­cien­cies to ex­ploit.” Just over the ho­ri­zon, they say, is “ad­dress­able” TV ad­vert­ising, which is already be­ing tested on some cable sys­tems. By 2016, it’s pos­sible that you and your neigh­bor may be watch­ing the same show but re­ceiv­ing dif­fer­ent com­mer­cials based on your demo­graph­ics. “If you have your tar­get audi­ence down to the in­di­vidu­al level, it un­locks some very power­ful mar­ket­ing tools,” McLean says.

(Photo Illustration by Jenny Mazer. Images: Adam Berry/Getty Images, left; Andy Jacobsohn, right) Photo Illustration by Jenny Mazer, images: Getty Images/Adam Berry, left, Andy Jacobsohn, right

(Photo Il­lus­tra­tion by Jenny Mazer. Im­ages: Adam Berry/Getty Im­ages, left; Andy Jac­ob­sohn, right)The tech­niques that AMG uses to place TV com­mer­cials more ef­fi­ciently are ad­ap­ted from di­git­al ad­vert­ising, where tar­get­ing can be per­formed with far more pre­ci­sion. In re­cent years, as con­sumer activ­ity has moved on­line, many com­mer­cial brands have shif­ted a third or more of their mar­ket­ing dol­lars to di­git­al me­dia. But polit­ics has been slow to fol­low suit. The Obama cam­paign was com­par­at­ively ag­gress­ive, spend­ing about 20 per­cent of its $580 mil­lion ad­vert­ising budget on­line. Jim Walsh of the Demo­crat­ic ad-tech firm DSPolit­ic­al pre­dicts that the trend will es­cal­ate in 2016, as cam­paigns seek to turn all the know­ledge they are gath­er­ing on in­di­vidu­al voter pref­er­ences in­to cus­tom­ized mes­sages. “The amount of dol­lars go­ing in­to di­git­al ad­vert­ising for tar­get­ing,” he says, “is go­ing to cre­ate a massive war.” 

DSPolit­ic­al bills it­self as the “home of the polit­ic­al cook­ie.” As most savvy com­puter users know, cook­ies are un­ob­trus­ive hom­ing beacons that your Web browser picks up as it tra­verses the In­ter­net. If you re­cently looked at a pair of shoes on­line and now see shoe ads every­where, you’ve seen the tech­no­logy at work. Walsh says DSPolit­ic­al main­tains a pool of around 600 mil­lion cook­ies and mo­bile-device pro­files. If one matches up to a polit­ic­al cli­ent’s list of tar­gets, it can be served an ad. Walsh likens the ef­fect of Web ad­vert­ising to the bill­boards that tout South of the Bor­der, a tour­ist trap on I-95 in the Car­o­li­nas. “For 100 miles, ima­gine you, your wife, and your two kids in the back­seat, all see­ing the same bill­board every 15 feet,” he says, “but for each of you in the car you’re see­ing a totally dif­fer­ent mes­sage that’s lit­er­ally tar­get­ing that in­di­vidu­al per­son.” You may not no­tice the mes­sage the first time, or the second, but even­tu­ally it pen­et­rates.

This form of ad­vert­ising is known as dir­ect mar­ket­ing: It is meant to in­spire a spe­cif­ic ac­tion, such as pulling off the in­ter­state to buy cheap fire­works, or click­ing on a link to donate to a can­did­ate. Dir­ect mar­ket­ing has tra­di­tion­ally been re­garded as a dark al­ley off of Madis­on Av­en­ue — a grungy coun­ter­part to the more luc­rat­ive and glam­or­ous “brand” ad­vert­ising that view­ers are ac­cus­tomed to see­ing on tele­vi­sion. Brand ad­vert­ising is de­signed to make you think, “Obama: Hope” or “Chevy trucks: They’re like a rock,” and it has a nos­tal­gic tele­vi­sion series about its glory days. But dir­ect mar­ket­ing pretty much owns the In­ter­net. Google makes its bil­lions by host­ing tiny little ads for lawn mowers to people at the very mo­ment they are search­ing for a lawn mower.

Di­git­al ad­vert­ising firms use a sim­il­arly dir­ect ap­proach to mar­ket can­did­ates and ideo­lo­gies. “If you’re not us­ing this solu­tion,” says Rich Mas­ter­son, founder of the Re­pub­lic­an firm Cam­paign Grid, “then 90 per­cent of your ad money is wasted.”

The place­ment of Web ad­vert­ising is typ­ic­ally done via ad net­works that sell massive num­bers of Web im­pres­sions across many sites. One morn­ing in May, shortly be­fore this year’s primar­ies in Pennsylvania, I vis­ited Cam­paign Grid’s sub­urb­an Phil­adelphia of­fices, where Mas­ter­son ges­tured across a room of work­ers in cu­bicles. “It’s just like a Wall Street trad­ing floor,” he said. “Ex­cept in­stead of trad­ing pork bel­lies, they’re trad­ing audi­ences for 105 cam­paigns.” 

Mas­ter­son doesn’t have a polit­ic­al back­ground. His form­at­ive ex­per­i­ence is in di­git­al mar­ket­ing. He star­ted a com­pany called U.S. In­ter­act­ive that rose and crashed dur­ing the first dot-com boom. In 2008, though, he was think­ing about run­ning for Con­gress in Pennsylvania, and he dis­covered how cam­paigns typ­ic­ally op­er­ate. “The polit­ic­al con­sult­ant told me I had to spend $5 mil­lion on ad­vert­ising in three dif­fer­ent states,” Mas­ter­son says. 

A cam­paign could po­ten­tially track wheth­er its tar­geted voters had vis­ited polling places.

He looked for Web-based al­tern­at­ives but found a void. “I just said, this is ab­so­lute in­san­ity,” he re­calls. “I was really shocked that nobody had solved it, be­cause you’re talk­ing about bil­lions of dol­lars in polit­ic­al spend­ing.” So in­stead of run­ning for of­fice, Mas­ter­son got to­geth­er with a friend named Jeff Dit­tus, whose back­ground was in the dir­ect mar­ket­ing of products like au­diobooks and Thigh­mas­ters, and star­ted Cam­paign Grid. Soon, they were work­ing for can­did­ates such as Chris Christie as well as the RNC, which Mas­ter­son says shared data that the com­pany used to build its own pro­pri­et­ary voter file. Cam­paign Grid re­cently re­ceived a pat­ent for what Dit­tus calls the “holy grail,” a sys­tem that pur­ports to in­teg­rate the voter data with ad-buy­ing pro­grams to auto­mat­ic­ally place Web videos and ban­ner ads in front of the right voters.

Be­cause so much in­vent­ory is avail­able across the Web, di­git­al ads are dirt cheap, but meas­ur­ing their ef­fect­ive­ness is a tricky busi­ness. Those on the cre­at­ive side of ad­vert­ising tend to be dis­missive of gen­er­ic ban­ners, and their “click here!” crudity. “There is no doubt that tech­no­logy has brought ad­dress­able mes­saging to new levels of soph­ist­ic­a­tion,” says An­drew Es­sex, vice chair­man of Droga5, a di­git­al agency that spe­cial­izes in brand ad­vert­ising. “But if you’re still serving them something that is the equi­val­ent of a te­di­ous bump­er stick­er, it is go­ing to be dis­carded and ig­nored.” 

Droga5, which works primar­ily with com­pan­ies such as Kraft and Coke but also oc­ca­sion­ally does polit­ic­al work, re­cently made a vir­al video in which celebrit­ies’ moth­ers im­plored their kids to sign up for Obama­care. It’s been viewed more than 800,000 times. By con­trast, a Web ad buy of 2 mil­lion im­pres­sions that Cam­paign Grid tar­geted to sub­urb­an wo­men on be­half of a can­did­ate in Nevada res­ul­ted in just over 1,000 clicks — a rate of .05 per­cent. That minus­cule ra­tio is ac­tu­ally pretty much in line with in­dustry norms. 

Maryland voters in 2012. (Chip Somodevilla/Getty Images) Chip Somodevilla / Getty

Mary­land voters in 2012. (Chip So­mod­ev­illa/Getty Im­ages)Even those num­bers may not al­ways be what they seem, however, due to the im­pen­et­rable and some­times seedy eco­nomy of on­line ad net­works. The audi­ence re­search firm com­Score has es­tim­ated that 36 per­cent of all Web traffic is gen­er­ated by bot­nets, rais­ing the pos­sib­il­ity that many ad im­pres­sions are fraud­u­lent. In­vest­ig­a­tions by The New York Times and The Wall Street Journ­al have re­vealed that video ads some­times end up un­wit­tingly dis­played next to porn, or on un­scru­pu­lous web­sites that show mul­tiple videos at the size of a single pixel. Sev­er­al di­git­al cam­paign firms be­trayed their alarm by pub­lish­ing writ­ten replies to the re­ports, ar­guing that the graft merely demon­strated the need for ex­pert guid­ance.

Di­git­al ad­vert­ising firms ar­gue that meas­ur­ing clicks alone un­der­states the ef­fect of their mar­ket­ing mes­sages, be­cause a re­sponse isn’t al­ways im­me­di­ate. Some are ex­per­i­ment­ing with oth­er met­rics, such as track­ing the amount of time your mouse spends hov­er­ing over an ad. And even a tiny click-through rate can still gen­er­ate a good re­turn on in­vest­ment if you con­sider the many mil­lions of im­pres­sions that can be had for the cost of a single TV spot. A few years ago, Cam­paign Grid ran an at­tack on a Demo­crat­ic House mem­ber who had voted to fund a study that tested the ef­fects of co­caine on mon­keys. The ad — which fea­tured a scream­ing chimp and the slo­gan “Does Your Mon­key Need Re­hab?” — was de­ployed in some 14 mil­lion paid im­pres­sions over eight days just be­fore the elec­tion. Af­ter­ward, a poll found that 4 per­cent of the people who man­aged to re­call the mon­key mes­sage switched their vote to the Re­pub­lic­an.

These are mar­gin­al ef­fects, but, like many ad tech sales­men, Mas­ter­son ar­gues that the in­dustry has only be­gun to ex­plore the fron­ti­ers of dir­ect mar­ket­ing. “This device is pinging all the time,” he says, hold­ing up his cell phone. “It’s a pass­ive check-in.” Con­sumer ad­vert­ising firms are cur­rently de­vel­op­ing sys­tems that use GPS tele­metry to “geotar­get” con­sumers. A store can send you a coupon as you’re walk­ing by to lure you in­side, or a car com­pany can ad­vert­ise to every­one who walks in­to one of its com­pet­it­or’s show­rooms. Mas­ter­son fore­sees all sorts of po­ten­tial geotar­get­ing ap­plic­a­tions in polit­ics. For in­stance, a cam­paign might track how many of its tar­geted voters had ac­tu­ally vis­ited their polling places on Elec­tion Day.

“Polit­ic­al ad­vert­ising is like build­ing a brand based on some­body that you hate.”

VOTER RE­GIS­TRA­TION data can tell a cam­paign where you live, but mo­bile tech­no­logy has the po­ten­tial to re­veal how you live: where you go, what you do, with whom you in­ter­act. Still, there are lim­its to tech­no­logy’s power over the elect­or­ate. “It’s fine to un­der­stand wheth­er someone is up for grabs,” says Chaun­cey McLean. “But can any­thing I say to you change your mind? Be­cause, if not, it doesn’t mat­ter.”

It does not seem like a his­tor­ic­al co­in­cid­ence that polit­ic­al tac­tics have taken on a mi­cro­scop­ic fo­cus at a mo­ment when both parties are mired in ali­en­at­ing small­ness. Tar­get­ing tech­no­lo­gies tend to be most use­ful for mo­tiv­at­ing the con­ver­ted, aid­ing the la­bor­i­ous work of or­gan­iz­a­tion and turnout. When it comes to the more neb­u­lous com­pon­ent of polit­ics — per­suad­ing the un­de­cided — their ef­fect­ive­ness is less clear. Des­pite a cen­tury’s worth of psy­cho­lo­gic­al study, per­sua­sion re­mains something of a mys­tery, even to the ex­perts. “The stuff I am ex­tremely skep­tic­al of is this idea that we can turn data in­to ad cam­paigns and ma­gic­ally turn people in­to voters,” says Jim Gil­li­am. “That’s not real.”

Jim Fer­guson, a long-haired Tex­an, is a vet­er­an ad man who worked at firms like Young & Ru­bicam, where he came up with mem­or­able brand­ing such as “Beef: It’s What’s for Din­ner.” In 2012, he joined Mitt Rom­ney’s cam­paign as its cre­at­ive dir­ect­or. “I came from a tra­di­tion­al brand-build­ing back­ground,” Fer­guson says, “and one of the things we were al­ways taught was, you buy from someone that you like.” But he dis­covered that the philo­sophy of per­sua­sion in today’s cam­paigns is dif­fer­ent. “Polit­ic­al ad­vert­ising,” Fer­guson says, “is like build­ing a brand based on some­body that you hate.”

Mitt Romney greets workers at call center on November 6, 2012 (Justin Sullivan / Getty) Justin Sullivan / Getty

Mitt Rom­ney greets work­ers at call cen­ter on Novem­ber 6, 2012 (Justin Sul­li­van / Getty)Fer­guson told me that the Rom­ney cam­paign grew so pre­oc­cu­pied with tar­get­ing nar­row sliv­ers of the elect­or­ate that it nev­er pro­jec­ted a broadly ap­peal­ing brand iden­tity. The cam­paign con­cen­trated on mak­ing ads meant to dir­ectly ap­peal to base voters, such as the one that de­ployed an out-of-con­text line — “You didn’t build that” — from an Obama speech on in­fra­struc­ture, an at­tack Fer­guson now deems “such bull­shit.” The ads mo­tiv­ated the faith­ful, but they didn’t change enough minds to win the elec­tion.

“All the evolved tools and in­nov­a­tions in polit­ic­al com­mu­nic­a­tions are fab­ulous,” says Re­pub­lic­an me­dia ad­viser Mark McKin­non, “but they don’t mean squat un­less you’ve got a great mes­sage and a tal­en­ted mes­sen­ger.”

I figured that if any­one is really cap­able of us­ing pre­ci­sion tech­no­logy to per­suade voters, it is the team that worked for Obama. So one spring morn­ing, I at­ten­ded a re­union of sorts at the Man­hat­tan of­fice of Bully Pul­pit In­ter­act­ive. Some 2012 cam­paign alums were now with the firm, which handled Obama’s di­git­al me­dia, and oth­ers had gone to work for Google, on the sale side of ad­vert­ising. The meet­ing’s at­mo­sphere was chummy, but the busi­ness at hand was po­ten­tially dis­may­ing: the com­ing midterms.

As the group met around a con­ver­ted ping-pong table, An­drew Bleeker, Bully Pul­pit’s pres­id­ent, went over a long list of en­dangered con­gres­sion­al seats. “I can’t be­lieve Col­or­ado is go­ing to be a thing,” said Jason Rosen­baum, a bearded Google ex­ec­ut­ive who formerly worked for the Demo­crat­ic Sen­at­ori­al Cam­paign Com­mit­tee. But more battle­grounds meant buy­ing more ad­vert­ise­ments via Google. In ad­di­tion to work­ing for Demo­crat­ic cam­paigns, Bully Pul­pit is design­ing ad­vert­ising for big-spend­ing in­de­pend­ent com­mit­tees like Nex­t­Gen Cli­mate Ac­tion, the lib­er­al bil­lion­aire Tom Stey­er’s su­per PAC. The firm told Google’s team that it planned to make heavy use of tar­get­ing to turn out voters. “Our buys, ex­cept in a very few cases, will not be na­tion­al,” Bleeker said.

Mark Skid­more, a Bully Pul­pit part­ner wear­ing a blazer with a pock­et square and his hair in a faux hawk, asked about geotar­get­ing phones and scrap­ing serv­er data to track op­pos­i­tion ads. Bully Pul­pit is work­ing to de­vel­op pro­pri­et­ary tech­no­logy to mon­it­or wheth­er Web ads are ac­tu­ally seen, as op­posed to just clicked on, in or­der to bet­ter test the ef­fects of polit­ic­al mes­sages. “This cycle, I’m very bullish on test­ing,” Skid­more said, “look­ing ahead to 2016.” 

Over lunch after the Google meet­ing, Skid­more, who used to man­age mar­ket­ing cam­paigns for Nike, said the Wash­ing­ton-based Bully Pul­pit had opened the New York of­fice to be closer to the cut­ting edge of ad tech. The com­pany was par­tic­u­larly proud to show off some re­cent work it did for Stey­er’s su­per PAC, which has an­nounced it plans to spend $100 mil­lion to sup­port Demo­crats and to press its en­vir­on­ment­ally centered agenda. “The or­gan­iz­a­tion is in­tro­du­cing it­self,” Ben Cof­fey Clark, the Bully Pul­pit part­ner who handled the cam­paign, told me as he pre­pared for a con­fer­ence call with Nex­t­Gen, where he would show the res­ults of an ini­tial cam­paign. “What you’ll see is a small part of a very large op­er­a­tion on the per­sua­sion side, where we were in­tro­du­cing Nex­t­Gen to elites.”

Dur­ing the call, Clark said the cam­paign, timed to co­in­cide with the White House Cor­res­pond­ents’ Din­ner, had res­ul­ted in more than 11 mil­lion ad im­pres­sions. It in­cluded tar­geted page “takeovers” on Politico and The Wash­ing­ton Post and mo­bile ads that spe­cific­ally hit users geo­located on Cap­it­ol Hill, on K Street, and at the Wash­ing­ton Hilton. Clark dir­ec­ted the meet­ing’s at­ten­tion to one of the takeovers, which was meant to cast un­flat­ter­ing at­ten­tion on Re­pub­lic­an Sen. Marco Ru­bio’s sup­port for the Key­stone XL oil pipeline. The ad fea­tured an­im­a­tion of Ru­bio’s head split­ting in two to re­veal the slo­gan: “Sen­at­or Ru­bio: Don’t Get Taken for a Suck­er.” “The head split­ting open was very much done in the line of high-im­pact poster art,” Clark said. “That real eye-catch­ing font from the pro­pa­ganda artists of the ‘60s.”

The im­age was strik­ing, yes, but per­sua­sion re­mains more art than sci­ence. If you were already con­vinced that the Key­stone pipeline was an en­vir­on­ment­al out­rage, the ad might in­spire you to make a call to Con­gress, or send a Tweet, or turn out to vote for a Demo­crat. If you were un­sure of the mer­its, though, would an evoc­at­ive at­tack on Ru­bio change your mind? Con­vin­cing the un­com­mit­ted isn’t as easy as flip­ping a switch. That, per­haps thank­fully, re­mains bey­ond the grasp of tech­no­logy. Politi­cians still must con­tend with an old-fash­ioned prob­lem: It’s not enough to get in­side voters’ heads. You have to have something to say once you’re there.

An­drew Rice is a con­trib­ut­ing ed­it­or at New York Magazine.

This art­icle has been up­dated to in­clude a longer ver­sion of Jim Walsh’s ori­gin­al quote about see­ing the same bill­board every 15 feet. 


Welcome to National Journal!

You are currently accessing National Journal from IP access. Please login to access this feature. If you have any questions, please contact your Dedicated Advisor.