As marketers still struggle with mobile platforms and their inherent tracking and targeting weaknesses across the mobile Web, apps and competing operating systems, I am hearing more from vendors
about "post-cookie" approaches to the problem. In fact, some believe that the alternative models being considered for mobile ad targeting will ultimately help inform desktop platforms. Only time and
testing will tell which of these approaches will prove effective and whether something like a standardized approach will emerge to help settle a market that many media buyers still find daunting.
Figuring out which mobile data points are the most important to determine relevance will be at the center of this conversation. Because granular location is one of the core variables introduced by mobile, that has become a special focus this year.
One interesting alternative model comes from the new mobile platform Voltari. Actually, Voltari is anything but new. This is the longstanding carrier-service Motricity under a new name and form. The company wound down its operator services to Verizon and AT&T on June 30, has moved from Seattle to New York, and is now a digital media platform emphasizing mobile.
Taking some of the lessons it learned from helping carriers identify customers’ propensity to buy into services and content, Voltari rejects the notion that any one signal can be used to target and optimize campaigns effectively enough. “We are looking at four data points,” says CEO Rich Stalzer. “Location, time of day, content and device.” While simple on the surface, those four pieces of information about a user looking at a piece of content can render up to 40,000 attributes, as each can be mapped across other known demo, psychographic and behavioral profiles.
The real strength of the system, Stalzer argues, is not just in the intial profiling as much as in the machine learning that follows. “We look at every moment every ad is served -- about these types of ads working at this location at this time of day.”
CTO David Castillo adds that the system was grounded in lessons from working with carriers and seeing users' propensity for engaging with different kinds of content. “One algorithm wouldn’t solve it all. We needed to put a pipeline chain together. It isn’t built on segments but on 40,000 attributes, and each model has positive and negative weights," says Castillo. "The negative weighting means I will optimize away from these attributes, and the positive ones we will optimize toward. Then there is a model to see what publishers perform for what type of content. And then contextual relevance targeting becomes a biaser.”
The system is scoring campaigns on the fly in real time, and making bids in exchanges accordingly. “It may impact the bid if it predicts a lift,” says Castillo, “and then we’ll determine the impact on the bid price, and then we will know immediately if we did something useful if we get a response or not. Then we reevaluate the scorecards.”
The company claims to see a significant 3X to 4X increase in overall click-throughs using the approach. “We learned it is very good at short flight campaigns. We can jump on a propensity curve by bootstrapping lookalikes against our own data,” says Stalzer. “With the machine learning algorithms, we don’t have to burn through as many impressions to get to the target population.” Movie releases, for instance, have performed especially well because the algorithms find the audience quickly and the persistent optimization mitigates some of the drop-off in performance that tends to occur over time.
The search for new targeting and optimization mechanisms is only really beginning for the mobile world, where so many variables are introduced and traditional tracking is challenged. The level of complexity for buying into these systems spikes even higher as the data scientists have to model and weigh the impact of precise location against user behaviors, against physical and content context, as well as time of day and a host of other possible signals. Welcome to the machine… mobilized.