Networks want to build audience segments to optimize tune-in campaigns, to reach the audience of a specific show or genre online, or to aid scheduling and show planning. These all lend themselves to different segment creation methodologies, as do the size and make-up of a network’s audience, and the number and types of shows they are segmenting.
How best to build segments is really both a data science question and a business question. In this post, we’re going to tackle why it might make sense, from a business perspective, to use one methodology over another – or at least to weight your algorithm more toward one of these parameters.
Give me reach
If you want to create large segments that aid reach in monetizing cookies or on-network retargeting targeting, then you should be using our total reach. This approach creates a segment of viewers that have watched a show for a given amount of time. Of all methodologies, this is the one with the lowest bar for a viewer to be included. I may only have had to watch half an episode of Real Housewives Of New Jersey to be in a segment of viewers who this show reached.
This creates large audiences, which lend themselves to any use case around monetizing audiences – such as matching your TVs and set-top box audience to online cookies – or where a network is rewarded based on the size of the segment it delivers. We’ve made the argument that segments can be problematic from a quality point of view – but for any advertiser looking for reach, this is the best way to do it.
This method is also great at targeting viewers on your network, but less so when modeling an audience on a 3rd party network. If you are building an audience using a lookalike model, you need attributes that you can use to find that audience elsewhere, but these qualities will be diluted if you use a reach-based method. Because the audience is so broad, you are likely to get common attributes that will make it difficult to identify this audience on other platforms.
Who watches the most?
As well as the quality issues associated with reach based segments, they also make it hard to compare different shows. It’s not valid to compare the reach of NCIS, which has multiple hour-long episodes of a given week, and a limited series like Baskets. NCIS is always going to have a higher reach because it airs more frequently. However, that doesn’t mean that people prefer it to Baskets.
Looking at those who watched more than the average viewer allows for more meaningful comparisons between a much broader range of shows. The same goes for popularity; this method will enable you to compare a niche show and a hit show more fairly.
The issue with using watched more than average is that it will over-index on viewers who watch a lot of TV. Let’s imagine we are looking for people who index heavily on two big shows – say, This Is Us and The Big Bang Theory. People who watch a lot of TV are very likely to watch both shows, and you will end up with a segment that is large, but probably not very distinctive. They may have watched these shows a lot, but they probably watched lots of other shows a lot as well.
This methodology produces better results if you are creating a mass market segment, and one that you think watches a lot of TV. Think older demographics, homemakers, etc. It works less well with harder-to-reach audiences, or for less favorite shows.
Finding the biggest fans
If you want to feed your segments into a lookalike model, for example, to sell TV audiences online, then you need to create the most distinctive set of TV viewers you can get.
You can do this by mining viewing data for the biggest fans of each show. This methodology produces smaller audiences but is another excellent filter for finding a show’s most loyal audience. It is particularly useful for creating segments of shows that may not be highly rated, but that have exceptionally loyal audiences, or for shows or channels that tend to attract lighter TV viewers.
It also accounts for viewers who may be heavy viewers of a specific show, but who are light TV viewers overall, more than the other methodologies. Inclusion in this segment is based purely on a viewer’s preference – and doesn’t take into account their overall viewing.
This methodology also creates the most distinctive segments of the four methods, and you will get much less overlap between segments, making this segment the most effective for inclusion in a lookalike model, or for fusion modeling.
We’d also use this approach to identify shows and programs which have small but loyal audiences. Any show with a large segment of viewers for whom it is a favorite show, as a proportion of the total audience, will be a significant driver of loyalty.