Forecasting Audiences: It’s Essential, But No One Said It Was Easy

By October 24, 2016 clypd Blog

Linear TV is a futures market: deals are agreed ahead of campaigns being aired, usually with an agreement that a particular audience level will be delivered. For this to work as efficiently as possible, good audience forecasts are essential. Where forecasts are wrong, the media owner either over delivers, effectively giving away audiences for free, or under-delivers, requiring additional inventory to be added to the campaign for free.

Forecasting is a lesson in humility because no one gets it right all the time (Exhibit A: the media’s current favorite geeky guru, Nate Silver, who failed to predict that Donald Trump would be the Republican presidential nominee). No forecast is perfect and no forecasting model is good for all time, because the landscape is always changing.

TV audience forecasts have to be made for all shows, including premieres, and while these new shows may have tested better than others before being put on the air, they may subsequently fail to bring the expected audience, either because expectations are surpassed (“it’s a big hit! We always knew it would be!”) or, sadly, not met and the show is canceled.

For a network that is trying to predict “business as usual” with its programs, the relative stability of audiences and share of viewers for programs and networks makes forecasting viable, but issues such as competitive activity, weather, seasonality, changing cable carriage agreements and viewing shifts to subscription VOD services all introduce varying levels of uncertainty.

And then there are unprecedented events. A great example of this is what happened with FXX in 2014. The chart below shows the Nielsen Live+3 average ratings for Adults 18+. It’s clear something big happened in August 2014, and thereafter the ratings shifted to a level about 50% higher than pre-August 2014 levels.


So what happened? If we drill deeper we see the shift in ratings beginning on August 21st. Some readers may recall what caused this shift: in 2013, FXX acquired broadcast syndication rights to “The Simpsons”. The FXX series premiere took place on August 21, 2014 with a 12 day marathon of all episodes broadcast in chronological order, plus The Simpsons Movie. With 522 episodes aired, this was the longest continuous marathon in the history of television, lasting 12 days. The first day of the marathon gave FXX its biggest ever ratings since launching in 2013 and these ratings continued to rise during the twelve days.

Did FXX know this would happen? They probably expected a lift in ratings, but they almost certainly did not know how high, and they were probably unable to predict the long-term improvement in ratings for the network that occurred.


Exceptions are always a problem with statistical models – a typical question asked of any TV audience prediction model is “but what about the Super Bowl?” The answer is usually something like “that would be treated separately…” That is a reasonable response for known events but not all are known – news events are not scripted. And events like FXX’s Simpsons marathon, while planned, are very difficult to forecast.

So forecasting is difficult but essential, and forecast models should be viewed as a journey rather than a destination, as the world changes every day and a forecast you made yesterday can be improved today. But let’s leave Homer Simpson with the last words: “Son, if you really want something in this life, you have to work for it. Now quiet! They’re about to announce the lottery numbers.”

Leave a Reply