A Better Measure Of Old Media?

Project Apollo aims to test the efficacy of TV and radio ads

There is a data gap in media today, and smart executives worry about it. Google (GOOG ) knows a ridiculous amount about what its searchers do and what they click on: It has terabytes of data on consumer behavior. TV networks and the marketers who advertise on them may have prettier ads. But when it comes to finding out what makes consumers jump, they sometimes must resort to guesswork, or at least the crunching of disparate findings from disparate sources.

Which is why radio-ratings firm Arbitron (ARB ) has for years been playing with what it grandly calls Project Apollo, a pilot test of which is nearing completion. Project Apollo is powered in large part by Arbitron's cell-phone-size Portable People Meter, which pings whenever a person carrying it is exposed to properly encoded radio or TV advertising, whether they're at home, in a bar, at a friend's house—wherever. The other half of the data is provided by Apollo partner ACNielsen's Homescan program, in which participating households use a different device to scan their product purchases into a database. Combine all that information, and you can shape findings to a well-honed geek-edge. You can learn, say, which commercials seem to turn occasional purchasers of a given product into steady customers.

Pulling all of this information together is, obviously, a gargantuan task. Studying some early findings of Project Apollo's pilot and talking with executives familiar with it show there are ways to generate and centralize some deep-diveable data from offline consumer behavior, but also that it's hard and costly to do so.

APOLLO IS BACKED BY SEVEN of the largest U.S. advertisers—including Procter & Gamble (PG ), Unilever (UN ), and Kraft (KFT )—which cumulatively spent over $7.6 billion on ads last year. The pilot test has been running since early 2005. A recent white paper presented by an executive involved with Project Apollo showed that applying its data to ad campaigns can make them 40% more efficient—that is, 40% cheaper. And a complex analysis of one unidentified household product's TV ads yielded the following insight: Those seen by its lightest users proved so wasteful that, once product sales and manufacturing costs are factored in, the campaign would have lost $24 million on those ads.

Dense stuff, I grant, but catnip for those who spend their days immersed in it. Still, even executives fond of Apollo express reservations. Don Gloeckler, P&G's manager of media research for North America (and—disclosure—the father of a BusinessWeek staffer), said some Project Apollo findings help the company "change the way we think about media." But he also noted "this data doesn't come for free. You need to make sure...you get a return on your investment." (One executive familiar with Project Apollo said that the companies involved in the pilot were initially asked to pay $1 million for the privilege. Cutting the pilot's participating households to 5,000 reduced the cost to $350,000.)

Apollo has limitations. The project works best with frequently bought items—more purchases make more data—which rules out participation from certain major advertisers such as car companies. Some media, such as print, do not lend themselves to people-meter measurements. Measuring them will require consumers to fill out surveys, which layers in additional complexity and the possibility of human error. An ad executive well acquainted with Project Apollo echoes Gloeckler's concerns, citing "expensive" costs for participants and the fact that the return on investment "is not crystal-clear."

This fall the seven companies will decide whether to go forward with the experiment or not. Like all ambitious new ventures, Project Apollo will require fine-tuning and deep pockets to make a go of it. And Google's database? In the time you read this column, it got even bigger.

For Jon Fine's blog on media and advertising, go to www.businessweek.com/innovate/FineOnMedia

By Jon Fine

Before it's here, it's on the Bloomberg Terminal.