Inside a Data-Driven TV Ad Buy
The set-top box data-fueled television ad trend is still emerging, and there are no shortage of young firms getting in on the ground floor. But is it really all it's cracked up to be?
Unlike the growing lot of startups claiming their secret data insights sauce is better than the next guy's, Seattle's PrecisionDemand aims to offer some proof. The firm recently revealed details of a series of tests it claims prove machine-based TV buys are smarter than human-determined ones.
There's nothing new about a TV targeting company touting the benefits of data-centric ad buying. Yet, it's not often that such firms pull back the covers on the algorithms they've developed, or present direct comparisons of campaigns planned with their modern methods to traditional TV audience targeting.
In one brand campaign comparison for a music subscription service, PrecisionDemand's number crunching led to more than 4,000 more subscribers than a previous campaign using a traditional media planning approach -- and it cost less than a third of the original campaign's price tag.
The company compared TV show buys chosen according to things like audience age and gender to targets selected using analysis of set top box data tied to advertiser's data on their customers. The system creates demographic profiles for audience targets and seeks TV shows that match best at the best price.
Results of four TV campaigns comprised of more than 22,000 airings were used in the study. PrecisionDemand recently published its findings in a geeky paper entitled, "A High-Dimensional Set Top Box Ad Targeting Algorithm including Experimental Comparisons to Traditional TV Algorithms." The paper will be presented at the IEEE International Conference on Data Mining in Dallas next month and is available at PrecisionDemand.com.
"TV is really kind of obsessed with impressions and age/gender breaks," said Brendan Kitts, chief scientist at PrecisionDemand, suggesting that this standard "lingua franca" of TV buying is about to be replaced by more targeted methods, similar to digital ad targeting. Mr. Kitts previously spent five years at Microsoft working on its ad serving technology.
For one thing, set top box data-aimed advertising promises to uncover less-than-obvious programming that is viewed by target audiences. In one test the company scored the relevance of TV shows to the target audience for a "Handyman product" ad on a scale of 1 to 10. In the end, some programs deemed inappropriate according to traditional standards were ranked highly by the PrecisionDemand system. For instance, shows including "The Nanny," "Inside Edition" and "The Fresh Prince of Bel-Air" got low scores from humans, but high scores from the machine. Another seemingly odd choice the system liked but humans ignored: Women's College Volleyball.
"The Media Buyers believed that this would not be relevant, however we think that handymen might actually like watching women's volleyball," states the report.
By surfacing unlikely TV show matches from the otherwise-neglected media options, set top box data-driven systems like these tend to create a larger pool of possible targeted inventory. "The amount of viewing behavior increases 800 fold from 25,000 to 20 million persons," notes the report.
Another ad trial for a digital music service aimed to reach young adults. The traditional way led media buyers to buy shows reaching 18-25 year old students. In that effort, cable channels including Spike TV, MTV and Comedy Central dominated the buy, together representing nearly 65% of the plan. The algorithmic approach created a far more diverse plan. According to the report, 823 distinct programs on 30 cable nets were purchased compared to 200 on 11 stations in the traditional campaign. The study also found that the modern targeting reduced cost per impression by 51%.