With The Weather Company now a unit of IBM, RTBlog recently spoke with Jeremy Hlavacek, VP for Global Automated Monetization, about the business. Hlavacek runs Weather’s global programmatic business and the SMB group.
RTBlog: What are the plans for integrations between Weather and IBM?
Jeremy Hlavacek: We’re still developing them. But we’re looking hard at location and weather data. IBM was attracted to all that data. And we’re learning more about the capabilities of Watson—how to plug it into our programmatic business. We’re still figuring out how it’s going to work. I would say we’re at the beginning of the integration process.
RTBlog:What are some of the programs you've done so far?
Hlavacek: We had one Watson ads product where you use the intelligence of Watson to inform the ad experience. We created banner ads that you can interact with. Watson is very good at voice and natural language recognition. It can recognize human language, and it’s being built into advertising that’s sold by The Weather Company.
We’ve run some pilot tests with Unilever, Campbell, and GSK Consumer Healthcare.
We’re hoping to learn more about how to improve the metrics and performance of advertising. For example, when you can talk to an ad like Flonase, you might want to know whether the medication is safe for kids. Watson can understand that and come back with an answer.
For Campbell, the question might be recipe-related, and Watson comes back with answers. We want to know how consumers will react to this. The hope is that Watson ads can fill a need.
RTBlog: What’s going on from a programmatic perspective?
Hlavacek: From a programmatic standpoint, the ability to create better ad products like Watson Ads will be valuable to our ad sales business. These are unique and valuable products.
Today, these products aren’t available through the programmatic channel; we’re going out first with direct sales of these products. Next year, they’ll be integrated with programmatic offerings. This requires custom integrations. And native executions, especially, are hard to execute programmatically.
My team spends a lot of time optimizing our existing programmatic ad stack. We try to extract the most yield, get good ad quality, and fill inventory. We want to see if Watson can help us understand our data better, improve our yield, traffic, and the bidding patterns from our partners. Adding the power of Watson to that equation could be a game-changer.
RTBlog: Why is programmatic mobile monetization so hard?
Hlavacek: At Weather, where I was for three and a half years, we had 70% desktop and 30% mobile, now it’s more like 50/50 and becoming more mobile than desktop. Consumers want weather data on their mobile phones.
It’s a challenge to monetize because the CPMs for mobile inventory are a lot lower than for Web inventory. We’re continuing to try to close that gap. Maybe eventually it will be more like 30% desktop and 70% mobile, but the impressions are currently worth less.
RTBlog:Why are mobile impressions worth less?
Hlavacek: The data on mobile users that is available works in a cookie-based marketplace. Retargeting is done to determine which impressions to buy. In-app inventory doesn’t rely on cookies. If you’re an ad buyer, and all impressions are generated in-app, you can’t tell which impressions are valuable to you and which aren’t.
On the flip side, there are some companies that are using probabilistic data to help marketers understand the data around these impressions. We need to convince marketers to spend their dollars in mobile channels. They’re skeptical because the data isn’t available, nor the in-app inventory.
RTBlog:How do you tackle this issue?
Hlavacek: You can use deterministic data, which means you have someone’s registration and login information. And you can tell who someone is, their demographic information, etc. Facebook and Google are really good at that.
We have 180 million downloads of the Weather app, and we have an audience, but we don’t have Facebook’s data and log-ins. We’re starting to build first-party data capabiltiies. We’ve invited certain users to share their names and information with us. Then we can ask them if they want a newsletter, and this captures a home location. As we build that dataset, our inventory will become more valuable to marketers.
Drawbridge and Tapad analyze the available data from mobile phone users, making their best guess as to who they are in a probabilistic model. Facebook uses more of a deterministic model.
RTBlog: How does weather affects programmatic media?
Hlavacek: Hurricane season began on Aug. 1 and when there’s a hurricane we see incredible spikes in traffic. It’s not unusual for us to see a 500% increase in our daily traffic when there’s a weather event.
It’s hard to plan for. Our yield department doesn’t quite know when an event will change the inventory mix.
But with programmatic, we can help sell the impressions that aren’t planned for. For example, if we wake up and find another 1 billion impressions we can use in programmatic marketplaces, it’s good for buyers. When we have a weather event, we have lots of new users on our site that might be valuable to our advertisers.