Last year we took a look at children’s television on streaming and how, since the COPPA (Children's Online Privacy Protection Act) protections that governed it forbade certain types of advertising, it was not being bought by the same teams that bought open programmatic inventory on streaming, but rather, was being sold directly. This left a gap, because agency TV buyers did not traditionally sell streaming and programmatic buyers did not traditionally buy direct. And that meant children’s programming on streaming was often undervalued.

This year, GumGum did a study on children’s television on streaming and found a far more disturbing conclusion: as much as 20 percent of the inventory on children’s streaming was not-COPPA compliant, meaning it was aimed at adults. Everything from liquor ads and casino ads on out. 

I sat down with Ken Weiner, Chief Technology Officer at GumGum whose clients include Dentsu, IPG, Publicis, Omnicom, WPP and Havas, to get a better sense of how they discovered this problem, and, more importantly, how it came about. Richie Hyden, co-founder & COO of IRIS.TV, joined me on the call.

ALAN WOLK: Let’s start off with the big story here. How exactly did you discover that all those adult ads were showing up on children’s streaming?

KEN WEINER: Our company is all about understanding context. And given that streaming can sometimes feel like the Wild West, especially smaller streaming services, we had a feeling that there were some examples of mistargeting going on and we wanted to bring those to light.

It doesn’t have to be kids’ content either—it can be a mismatch between the types of content and the ads they’re showing, like putting a Toyota ad into a car crash scene.  

ALAN WOLK: How does that happen though, that adult ads get shown during children’s programming? Because unlike the Toyota example, there’s an actual law that’s supposed to prevent that. 

KEN WEINER:  It happens if people aren't actively trying to prevent it from happening and using the right tools available to do so. There are certain guidelines that are set up for content that's made for kids where it is supposed to be marked as meant for kids. And then advertisers are supposed to look for that signal and steer away from the content if their advertisements are not COPPA-compliant.

So if one of those players fails to indicate that there could be a COPPA guideline flag, that information might not make it up to the buying platform. In which case there's nothing the buying platform can do because they don't know what they're buying. 

ALAN WOLK: What is the solution then, how can we ensure this doesn’t happen?

KEN WEINER: GumGum has a product called Verity that is designed to prevent situations like this. 

To start, you have to assume that no one did any of the tagging on the content properly. 

At a very basic level, there are some contextual categories that are part of a taxonomy we use to identify content which would indicate that there's a likelihood that the content in question is made for kids. It can’t tell you definitively, but it could identify things like “this is a cartoon” or it can make assumptions based on the dialog as part of a scene-by-scene analysis of a video. 

So once those child-related taxonomy nodes are identified, advertisers with COPPA noncompliant products could make sure their products do not run on those shows. 

ALAN WOLK: Does this type of analysis have to be done every time the show airs?

RICHIE HYDEN: No you do not need to run this type of analysis every time the piece of video programming airs because the content doesn’t change. What is needed is a process for unlocking video-level data, which is a deep understanding of the piece of video content playing on glass that can only be derived from watching the piece of content. 

For video-level analysis to work at scale, you need a common content signal to enable solutions to enrich the video-level data (currencies to the signal if you will) and buying platforms to activate on it. In addition to having their own tagging policies, publishers are concerned about data leakage, cherry-picking, and privacy. We developed the IRIS_ID as the content signal for publishers to securely share video-level data. We integrate with their content management system (CMS), ingest and normalize all their video-level data including video URL and program-level information and assign each video an IRIS_ID.

We send this normalized data to data partners like Verity for analysis and enrichment. All this enriched data is then associated with the IRIS_ID for that particular video which is then passed into the bid stream where it can be decoded for ad targeting across direct, PMP, and open auction deals.

KEN WEINER: So what IRIS.TV solves for is giving each video an identity by essentially assigning a universal Content ID to it that they call the IRIS_ID.They then send the video and its IRIS_ID to companies like us to analyze. We can then attach our contextual understanding of the video to therefore the IRIS_ID. Once this is completed, the IRIS_ID is passed in the bid-request  to ad platforms (ad servers, SSPs, DSPs)  where it can be decode to understand the Verity definitions or segments for that video. This recreates the workflow we have for understanding what the text on a web page is but receiving a page URL and understanding the contextual information of that page using Verity.

So that's how IRIS.TV helps, by giving a common and standardized identity to all the videos that are out there that are passing through the ad tech ecosystem to enable this type of targeting.

ALAN WOLK: How much awareness is there that this is an issue, that a lack of any kind of standardization around shows and their context leads to problems like gambling ads showing up on kids TV shows?

KEN WEINER: I don’t think many advertisers realize how much of an issue it is to not have a unique content identifier like the IRIS_ID across the inventory they are buying. People are coming at video buying from two different angles.You have companies that were doing display advertising on the web and are now trying to get into video and you have people who were in TV advertising and now they’re trying to get into streaming, which is bought and sold like digital.

They have different expectations. Digital companies demand a lot of performance metrics and reporting and transparency. They're used to it. The ones that are coming from TV, they're not used to having all that.

But overall, I think they both are looking at audience buying as the primary way to buy streaming TV.  So they don’t appreciate how much can be done with contextual buying and also what the longevity of that is versus audience buying, which is continuously at risk and is diminishing every month.

ALAN WOLK: So how do you see all that shaking out?

KEN WEINER: As all the data privacy demands start to come to fruition it’s going to be a problem on streaming too. So everyone is going to be looking for solutions that don't involve user targeting. That's where I think the interest in contextual and brand safety is going to grow and grow.

That’s why we’re looking to improve our platform so we can tell you context down to the moment. Because in a two hour movie that’s shown on TV, there are going to be funny scenes and sad scenes, scenes about sports and scenes about cooking. And advertisers are going to want to know all of that, including what is happening in the scenes right before their ad runs and in the scenes right after.

RICHIE HYDEN: Despite privacy regulations, people-based signals in streaming cannot be accurate. This is especially the case in CTV as there are multiple people in a home watching on smart TVs from different manufacturers. The reality is that content signals are identity signals and the best way to target users in relevant and brand-suitable experiences is to show the advertising that is contextually aligned to the types of programming they are watching in that moment in time. We have seen that mood and mindset drive higher campaign performance than demo targeting or even being in the market for a specific product.