Nielsen Suggests ‘Fusion’ Model To Better Track ROI

At a closed-door meeting last week with TV networks, ad agencies and other clients, Nielsen executives for the first time proposed using a statistical “fusion” model to measure product purchases in Nielsen TV households.

The statistical model, designed to address advertisers’ increasing demands for better ROI measurement tools, would allow Nielsen to merge data collected from the AC Nielsen Homescan service with the regular TV ratings produced by sister company Nielsen Media Research (both are owned by Adweek parent VNU). If the plan works and clients sign on, Nielsen said it would compete more directly with third-party data processors, such as Simmons and MRI, that provide an array of consumer behavioral and lifestyle information based on audience research.

While agency and network executives last week applauded Nielsen’s efforts, they cautioned that they didn’t want the monopolistic ratings company to limit access to its raw data to third-party research providers. “It’s certainly something worthwhile to look at, and hopefully Nielsen doesn’t use it to cut others out of the pro-cess,” said Bruce Goerlich, evp, director of strate- gic resources at Publicis’ Zenith Media. That is “absolutely not” the intent, said Nielsen chief research officer Paul Donato.

Clients said the proposal, made by Nielsen vp, analytics and modeling, Pete Doe, has merit, but also leaves many unanswered questions. Chief among them is whether testing will prove Nielsen’s claim that it can merge the two databases with fusion models. Issues such as cost, flexibility to meet individual client needs and predictability of the data haven’t been addressed, sources at the meeting said. Donato says much of the necessary testing is complete, and that “validation data” will be presented to clients in the coming months.

In effect, the proposed service would be a stripped-down version of Project Apollo, which is currently in development by Nielsen parent VNU and Arbitron, and some Nielsen clients wondered if the ratings company was trying to offer an affordable alternative, should Apollo not move forward. Though many ad and media researchers think highly of Apollo, it is expensive, and after a year, the companies have not secured enough funding [see Adweek, Oct. 3].

But Howard Shimmel, Nielsen’s svp, client insights, said, “These are two different spaces, and we are not doing this as a hedge against Apollo. We believe in the Apollo concept.”

Researchers say that fusion techniques, common in Europe for years, were rejected until recently in the U.S. because such techniques use mathematical models that calculate likely behavior—in this case product purchases by TV viewers—in lieu of measuring actual behavior.

But more sophisticated modeling techniques have eased those concerns, according to researchers. Still, the preferred technique is to have a single source of data, like Apollo, where the same consumer panel that’s watching TV, listening to radio and consuming other measured media is also doing the product purchasing.

“It’s good to see fusion on [Nielsen’s] agenda,” said Jim Kite, evp, director of insights, research and accountability at Publicis’ MediaVest. The agency does fusion modeling in-house for many clients. “There are so many different sources of information, and we have to get them talking to one another.”

But fusion still has its detractors, including Apollo partner Arbitron. Company CEO Steve Morris, talking to analysts last week, called fusion “a failed idea,” and predicted there would be little or no demand for the kind of service Nielsen proposed.

Others wonder whether it’s possible to estimate future viewing based on product usage. “People don’t necessarily watch TV programs because they use Tide,” said Susan Nathan, svp, director of media research at Interpublic Group’s Universal McCann. But, she’s intrigued by Nielsen’s proposal and wants to know more.

Tim Brooks, svp, research, Lifetime Television, says there are “lots of questions” about the reliability of Homescan data, such as how representative of the overall population it is. “The data that is fused is only as good as the data that underlies it,” he said.