If measuring TV audiences is getting more and more difficult, accurately forecasting the level of viewing beforehand in order to sell advertising is even trickier.
SintecMedia, an advertising management software company that is part of a consortium of business, government and academic data science experts started by the Israel Science Ministry, said it has created an advanced machine learning algorithm that predicts viewership of TV programs.
The algorithm has been tested with what the company describes as a major TV media network in the U.S. and produced better results than the manual process the network now uses, said SintecMedia President Lorne Brown. “Now we’re working on a long-term contract with this network.”
SintecMedia believes its results will improve as testing continues and the machine learning continues to fine-tune its results.
Predicting TV viewership is important to networks because they sell ads based on the expected audience. They also guarantee that the commercials will deliver that number of viewers.
When the shows fall short of expectations, networks must provide makegoods to advertisers. The networks leave money on the table when shows deliver more viewers than expected.
SintecMedia estimates that TV networks lose up to 10% of their ad revenue because of inaccurate ratings projections.
“We decided this was a big problem that needed to be addressed,” Brown said.
The SintecMedia consortium—as part of a five-year, $25 million initiative—created a long and short memory network that predicts gross ratings points. It uses neural networks and natural language processing.
Most networks base their ratings estimates on the past performance of the show or related shows in similar timeslots.
SintecMedia also uses historical performance data but looks at the content of the shows, based on scripts, in order to determine how popular they’re likely to be. It also incorporates social media data from Twitter and Facebook, trending topics on search, weather and holiday data.
“These type of predictions that we are making are very unique, and no one is approaching it this way whatsoever,” Brown said. “This is a way to give more control to media companies. As our road map goes further, this is going to be important for predicting addressable TV and streaming video, live TV via TV Everywhere or over the top.”
At this point, SintecMedia is making estimates based on the standard Nielsen demographic groups now used to buy and sell most TV advertising.
But the company says that as TV companies move toward using data to sell based on more specific consumer targets—people who intend to buy SUVs, for example—its algorithm will be able to incorporate additional data to create estimates for those audiences as well.
“We’re going to use that same algorithm that looks at show content as well as social media to determine how you forecast that as well,” Brown said.
Companies that do that type of audience selling, including NBCUniversal, Turner Broadcasting and Viacom, have their own predictive algorithms.
“This would be fully complementary and help them.” Brown said.
SintecMedia says its media management software is used by 300 companies, including NBCU, CBS, ABC, AT&T and Starz.