Big Data: Only Half of the Marketing Equation

Recently, we covered the basics on predictive marketing, the closest thing a marketing department has to a crystal ball: it can help you decide how to provide the goods and services consumers want, even before they ask for them.

How do you do this? By using propensity models — analytics that are a gold mine of consumer data. Analyze consumers’ past behaviors to hone in on your target demographics and figure out the best ways to reach them. Think about the way Netflix offers a list of recommended shows, or how Amazon offers a list of recommended products to buy.

But here’s the rub: Propensity models can only work with known data. What they don’t account for is the unknown.

Here’s a good example of predictive modeling gone awry: Pinterest, the website and app that allows users to create digital inspiration boards, has been using propensity models for a long time. Every so often, Pinterest will email its users with “recommended boards” and “recommended photos” based on what those users have been pinning lately. If a user has been pinning a lot about food, Pinterest can recommend a few photos of gourmet Italian meals. It’s a way to generate continued interest in the site, as well as introduce its user base to new features on the website.

But last month, Pinterest got caught in a predictive modeling snafu. The site sent hundreds of women emails with the subject: “You’re Getting Married!” with an enclosed list of recommended wedding inspirational photos, after noticing that they had been pinning a lot to their wedding boards. Here’s the thing, though: Although the consumers had all been focusing on wedding-related content, they weren’t getting married. They were just pinning pictures they liked.

The problem? Pinterest didn’t understand its user base. To any twentysomething woman who uses the site, it would seem obvious that many Pinterest boards are purely aspirational—and clearly not every woman with a wedding board is a bride-to-be. But whoever was analyzing Pinterest’s user data didn’t make the connection between how the boards were being used and what they revealed about the users. The moral of the story is to put yourself in your customers’ shoes, and really get in their brain before you analyze the data.

Predictive marketing can do wonders, but data is only half the story. You have to know and understand your consumer base before you can capitalize on past behaviors. A propensity model might tell you what consumers have done in the past, but it’s up to you to make the connections to what that means for your company in the future.