“On TV & Video” is a column exploring opportunities and challenges in advanced TV and video. Today’s column is by AdExchanger Sr. Editor James Hercher.
It wasn’t so long ago that around one-fifth or more of the entire adult-age population of America would simultaneously sit down and watch a TV show at the same time.
In such a media environment, a TV panel like Nielsen’s was a useful currency.
But now, content consumption is split not just across hundreds of individual TV networks, but across video-on-demand services, smart TV operating systems, CTV apps, mobile devices, browsers and more.
“With such fractured and granular viewing habits, panels are just not fit for purpose anymore,” said Josh Chasin, chief measurability officer of the video advertising and analytics company VideoAmp.
Panels do still serve a purpose, he said. They can be used to crack open the household-level reporting data that TV measurement services get from TV networks, set-top boxes and streaming platforms like Roku or Hulu.
Panels are also helpful for answering questions about how content is consumed within households. Are certain shows viewed by a group within the household, or is it one individual watching alone? Are the adults watching the big screen on the wall while teenagers stream the same content to their phones and laptops?
AdExchanger caught up with Chasin to get into the pros and cons of panel data for advanced TV and video advertising – and how non-Nielsen panel data sources are filling gaps in the market.
AdExchanger: Why do we still need TV panel data?
JOSH CHASIN: It’s generally accepted now that audience measurement has become a science that requires accessing and aggregating big data assets. Those data sets for TV and CTV generally include set-top box data, smart TV automated content recognition data and digital log file data or digital data collected from pixels.
Then you have large and useful – but imperfect – data sets. Panels are still necessary in this framework, because for certain things you can treat panels as a truth set. You can use them to weight, adjust or calibrate the data.
To be clear, when I say a panel, I mean a pool of households and that the people in those households are recruited for a purpose and consent to have certain components of their behavior tracked, as well as to report on their demographics and ways to classify that data.
What do you mean by “treat panels as a truth set”?
What panels do is enable us to modify big data assets based on accounts or get to individual-level data based on what we know about who’s in the household.
If you want to know about women ages 25 to 54 in our data, it will be households with a woman who’s in that range. The next step is to convert household viewing data to individual-person viewing data, and the way you do that is with a panel that informs you how the people in households watch a given show or channel.
My household includes myself, my wife and my daughter. At the household level, you wouldn’t know we were watching, let’s say, “Modern Family,” but a panel may know that the Chasin household is watching it. And when I say that panels can “personify” the big data set, I mean it can identify that three people are watching the same show – an adult male, adult female and teenage female. One of the things this does is let us turn a household impression, which by definition becomes one single impression, into impressions that reach multiple people.
With the household-level data sets and the panel personifying the data, we may be able to put the information together to determine that a campaign that reached one million households also reached 1.3 million individuals and 700,000 of those were female.
Dealing with both buyers and sellers that aren’t simply relying on panels for linear ratings – that’s one of the main applications for panel data right now.
Everyone knows Nielsen – but what other panel operators are there?
There are two other companies in this space. One is TVision, the other is HyphaMetrics, and we work with both.
What are the differences in approach?
I want to be careful to stay agnostic here. But I’ll note that TVision has been around longer and they focus on attentiveness and engagement metrics. Their primary business model has been to help understand which programs and which ads have retained the attention of viewers. They measure eyes on set.
With HyphaMetrics, I’ve heard them use the term, “A panel for the rest of us,” by which they mean non-Nielsen panel data that can be used by companies like VideoAmp. Their panel data is to inform your own cross-channel ratings.
What other vendors or partnerships are important for your ratings information?
There are many data sources. One is set-top boxes. We work with Frontier, TiVo, Dish and have one or two more MVPDs coming. We also have smart TV data. Last month we renewed a data partnership with Vizio and we are one of very few companies still in their data-sharing program. That’s important data because they can fingerprint the screen to know what ran and the content being consumed.
There’s server log file data, too, which we acquire through arrangements with TV networks that provide us with access to their server logs.
Why is that valuable?
Say a broadcaster, for example, agrees to share data with us. We may see streaming data through a CTV app combined with linear TV data in a privacy-compliant fashion to be able to predict the number and kinds of audiences that saw an ad streaming compared to the linear ratings. That data sharing could also include pixel-based data, which might help us understand the subset of content viewed by phone, by laptop or smart TV, etcetera.
At that point you have what we would call a commingled footprint, with smart TV and set-top box data sources mapped to some millions of households. We use LiveRamp and Experian to map devices to those households.
To do all that in a way that’s effective and also preserves the broadcaster’s proprietary data, we work with a clean room data environment.
And that’s about as far as we can go. It’s the same for everybody that’s doing this without a panel.
This interview has been edited and condensed.
