At Disney, no one messes with the Mouse.
The brand is extremely protective of its reputation among consumers, one largely built on trust.
At the same time, Disney is a data-driven publisher with an epic amount of O&O. There’s ABC, ESPN, Fox, Hulu, movies, hotel resorts, cruises, theme parks and everything else that formed the foundation of your childhood – and the company has taken great pains to ensure advertisers can message across its entire portfolio.
Ad president Rita Ferro’s promotion last year to oversee all of Disney’s ad sales catalyzed a movement to make the Mouse House’s inventory available and targetable.
But Disney is just as careful with its consumers and their data as it is with its brand perception – so much so that Disney takes a cautious and very specific approach when allowing its audiences to be messaged by advertisers.
Yes, data across all of Disney, including theme parks, can inform advertising. But you’re not going to be able to message somebody directly based on an action that they took at Disneyland or any other Disney property.
Instead, Disney uses machine learning algorithms to model how its audience segments are likely to act, and advertisers can target Disney consumers based on those predictions.
At Cannes, AdExchanger sat down with Dana McGraw, Disney’s VP of audience modeling and data science, to discuss how Disney creates its segments, the Disney identifier and the heady job she’ll have incorporating data from the properties Disney acquired from Fox.
AdExchanger: I’m guessing you’re working to integrate all of the Fox assets these days.
DANA MCGRAW: We are, with FX and NatGeo. I spend time figuring out what data is there, how does it look. But the more data, the better. We do a lot with predictive modeling of various segments.
Disney also has a controlling stake in Hulu. Is that data easier to integrate, since you’re not totally unfamiliar with it?
We know what the data looks like, because we’ve been inheriting it for some time, based on how our shows performed. That’s easier in the big picture, since we know how the data is structured.
With FX and NatGeo, it’s on the roadmap, we have a sense of when we’ll have that data cleaned and segments built and we’re moving on that as fast as humanly possible.
Did you ever get any data from Disney movies on Netflix?
None that I ever interacted with. I’m sure there was some top-line data, like how many views something got, and how much something was consumed. But in terms of the type of data you need to do any data science on? No.
You don’t want too many segments, or you lose reach. Is there an ideal ratio between getting reach and good targeting?
It really depends on audience behaviors. It depends on the psychographic and behavioral data, how homogenous your audience is and how much you can cut it to have broad enough reach within each segment – and accuracy.
We certify our segments in terms of the accuracy. We won’t sell a segment that’s 51% accurate, because the outcome won’t be good for the brand.
Do you have segments based on Disney properties like ESPN or NatGeo, or do they cross all of Disney’s portfolios?
We started as separate entities, even those owned by Walt Disney before. But the idea, as part of the upfront, is figuring out how to sell across platforms, across segments, while being as flexible as possible when someone has a buy with us, so we can move inventory around, but still find the same audience.
All of our segments are built to cut across everything.
Does that include theme parks and merchandise?
We get that question a lot. At the corporate level, we have a data management platform, where we all put in some amount of anonymized data across every division of the company. And we are able to leverage that data for buys.
How do you get other departments to put in that data?
Every piece of data certainly does not go in [the DMP]. There are some things that just need to stay with the [amusement parks] and we wouldn’t think of leveraging that. Same with each group.
But there is some amount of baseline data that, at an enterprise corporate level, we all agree needs to be in there.
How do you guys connect all of this? Do you have a Disney ID that spans between everything?
We built our models based off of device ID. I started with the Interactive Media Group, which spans everything now. We have a very large network of apps and games, much larger than most people anticipate.
Our job was to look at churn prediction and revenue optimization. We then realized we need to understand the audience and know more about who they are.
So, we worked with a mobile panel. We’d ask survey questions like if you were a man or a woman. Where did they fall in terms of perceptions of certain brands and of us? Were they in the market for a car? We were trying to do a basic audience profile, but then realized we can ask a lot more questions and be predictive of all kinds of things.
Then we do a blind match with the [device IDs from the mobile panel] to connect [the survey takers] to device IDs within our data repository.
We use machine learning models to make predictions and discover the positive and negative coefficients for behaviors that are predictive of, say, you being in the market for a Toyota.
Now we have all the data across the Disney/ABC properties and ESPN, and we can look across them and fit them to machine learning models and make more accurate predictions.
We have 140 million device IDs in the United States. Close to 400 million globally.
Do you go to market with a Disney ID?
We’re in market talking about our predictive modeling capabilities.
The Disney ID piece is separate and happens on the back end of the Walt Disney Company. For advertising purposes, we really key off of device ID.
But say we’re working with a brand that’s willing to do blind data matches. A brand may say, “Here’s people who are valuable to us.” We can map our two data sets and either hyper-target those people and build look-alikes, or maybe they want to reach new people who aren’t in their data set and we can suppress that group.
This interview has been edited for clarity.