4A’s CreateTech: With Predictive Marketing, Focus On The In-Store Action

Machine-learning technology can help marketers serve better messages to individuals — but trying to predict clicks can cause problems, says Dstillery’s Claudia Perlich.

Predictive marketing technology can help brands serve the right messages to consumers on an individual level — but brands that use the model to measure campaign success based on clicks won’t see results, warns Dstillery’s Claudia Perlich.

Claudia Perlich
Dstillery’s Claudia Perlich

In a session at the 4A’s CreateTech conference on Predictive Marketing: Promises and Pitfalls, Perlich broke down one of the key challenges in terms of making the most of machine-learning. The technology — which analyzes large swaths of past behavioral data in order to determine which people have a high propensity to buy product X in the future — learns to predict accidental clicks much more easily than intentional clicks.

Put simply, this happens because accidental clicks occur in easily predictable situations: Users of certain games, for example, reliably click ads while swiping a finger across the screen, and the model learns this behavior pattern. As a result, Perlich said, “when the measurement metric is clicks, [marketers] are basically targeting at random.”

The effective use of predictive marketing, Perlich suggested, is to improve targeting by determining how can impact real-world behaviors and purchases. The model can make predictions for a series of different types of ad creative, for example, and produce the percentage possibility that it will convert to an in-store purchase.

“It gives you the ability to say, ‘with all the information I have about this person, this ad is the best.’” In terms of analyzing the in-store outcome, Dstillery does this by matching shopping data with loyalty program Shopcomm, Perlich said.

“This is not an audience in the traditional sense,” Perlich continued. “It’s people the machine finds have a high propensity to take a certain action. As a result, it considers people as individuals — and has much higher results than generic when it comes to predicting if people will download an app, buy a product, or even move across the county.”

Essentially, the technology looks to be a way to simplify the process of data analysis and targeting, enabling marketers to cherry-pick individuals who have the highest propensity of taking a certain action — and then giving them a “nudge” to inspire the purchase.

“The bottom line is that consumers expect to be treated like individuals, so that’s where brands need to be,” Perlich said. “Bringing predictive modeling into the realm of machine learning to talk about people as individuals and not just “audiences” is a way to do that.”

About The Author
Lauryn Chamberlain Lauryn Chamberlain @laurynchamberla

Lauryn Chamberlain is the Associate Editor of A New York City based journalist, she specializes in stories related to retail, dining, hospitality, and travel.