Some marketers are disillusioned with data management platform technology, but Adobe is still actively investing in its DMP.
On Wednesday, Adobe added a feature to Audience Manager that allows marketers to associate unknown users with segments based on their propensity to take a certain action. The feature, which was in beta for roughly a month, is now generally available.
Around 50 marketers are already either testing or using predictive audiences in Audience Manager, including Sprint and the NFL.
The predictive audiences rollout is proof that Adobe isn’t giving up on its DMP, said Matt Skinner, senior manager for product marketing at Adobe.
“We still have hundreds of DMP customers,” he said. “Releasing a new feature like this in 2020 is evidence that we’re dedicated to enhancing our existing products.”
A good way to think about predictive audiences is as a complement to look-alike modeling. They’re like two sides of the same coin, Skinner said.
Look-alike models expand reach to unknown users with the qualities of existing users, while the predictive audience tool tries to match unknown users with known audience personas.
The primary use case for predictive audiences is to enable marketers to serve personalized experiences or messages to users they don’t know that well yet, such as first-time site visitors or someone who’s browsed but never bought anything.
Having gone through the beta and verified its audience output, Sprint is now gearing up to launch a live test. It will use predictive audiences to bucket its site visitors into segments based on specific traits and attributes associated with “high ROI” customers and “low ROI” customers, said Jeff Henshaw, VP of digital product analytics at T-Mobile.
From there, Sprint will be able to test serving different messaging and content based on a customer’s likely behavior.
“In the past, there was a lot of overlap or gray areas when we looked to build groups for potential high and low ROI customers,” Henshaw said. “With predictive audiences, we’re seeing more distinct traits.”
Predictive modeling is a key component of Sprint’s overarching and ongoing approach to digital transformation.
“We’re always testing different messaging and functionality,” Henshaw said. “[This] is another tool we can utilize that will allow us to continue to improve the digital experience for our customers.”
Many marketers still rely on assumptions when they’re building segments.
“I know this firsthand,” Skinner said. He spent three years as senior marketing manager for demand generation at Levi Strauss & Co. before joining Adobe in 2016.
“It’s often a best-guess scenario based on business needs and what a brand thinks it knows about its users,” he said. “We’re applying an algorithm to try and help marketers take advantage of more data-driven workflows.”
To build its predictive audiences, Adobe ingests a customer’s hashed first-party data, web analytics data, site tag data, pixel data from ad campaigns – whatever clients put in the DMP. A machine learning model uses the data to generate a propensity score for unknown users. Support for additional data sources is on the road map, Skinner said.
Although predictive audiences is a feature built within Audience Manager, brands that also use Adobe’s customer data platform can activate their predictive segments there too.