Event: Learn Data Science from Experts at Google and Netflix Register now

Online media companies typically generate the bulk of their revenue from ad impressions, but they also rely on ads to attract readers to content in the first place. With this business model, maximizing profit is all about efficiently allocating ad spend to attract the right audience. That’s why online media giant Topix takes the data-driven approach: The team leverages a predictive model, more specifically a Facebook ad classifier, to algorithmically determine how any given Facebook ad campaign will perform.

A classifier is a type of supervised machine learning algorithm that uses historical data to identify which set of categories, called “classes,” a new observation-- in this case, an ad campaign-- belongs to. The classes in Topix’s Facebook ad classifier are defined in terms of profitability: The goal is for the algorithm to recognize whether a specific ad campaign will be profitable or unprofitable in the long run. It works by analyzing patterns in historical data in which each campaign has already been correctly labeled as one class or the other. This step is called training the model.

Once the training period is complete, the classification algorithm can recognize which explanatory factors or “features” in the data reliably inform the class that each observation belongs to. In the case of Topix’s Facebook ad classifier, examples of key features that help definitively determine whether a campaign will be profitable or not include profit margin and cost per click. With these features as the model inputs, the classifier predicts each ad campaign’s profitability within the first two days of campaign testing. This allows the Topix team to avoid losing money and to quickly reallocate ad spend where it will have the greatest positive impact.

As a result, Topix observes two major gains from evaluating the performance of its ad campaigns algorithmically: time and money. Manually monitoring and adjusting hundreds of diversely targeted Facebook ad campaigns for each piece of content could take a week or more, but the company’s data-driven approach returns reliable results within two days. What’s more, hosting the Facebook ad classifier in the DataScience Cloud means the company’s data science team can easily iterate on the model and decision makers have access to its daily outputs.

To learn more about how the DataScience Cloud helps Topix make strategic decisions, download the case study.