By Michael Lamb, Director of Product Marketing, Verizon Media
As data-driven marketers, we are used to audience segments with specific labels, such as "football enthusiast" or "mothers with young children". While these labels are directionally helpful, they rarely represent the entire audiences' composition. While this might seem misleading, the best segments go beyond our expectations and take into account a diverse group of signals. They are able to uncover a complete picture of how consumers act online and how they respond to different products and services. So let's take a moment to go behind the scenes on the science of building audiences based on behavior.
Most behavioral segments today are sourced from online visitation patterns. As we consume content, the keywords and categorizations of that content become strong signals as to interest. But interests are not gender specific. A user who searches for information on diaper rashes and reads encouraging stories about singing to infants could be a mother or father. Another user who watches videos with beauty tips and reads hair care articles could be a teenage boy or girl. Our internal biases make assumptions about who should be consuming "male" content and who should be consuming "female" content.
There are a couple of techniques that help behavioral segments reach the people most likely to resonate with a brand message. The first is using multiple signals, such as search queries, emails, known online and offline purchases, as well as social signals and location trends to develop the segment. The more ways to "confirm" someone's interests, the more likely the behavioral segments will be accurate. The second aspect data providers must do is to constantly fine-tune the particular signals used to drive behavioral segments. Given the quantity and complexity of interests available, this process of refining segments requires systems dedicated to natural language processing, machine learning, and prediction algorithms. The quality of the segment, and therefore its performance, are often tied to the strength and quality of the data and algorithms used.
There's a specific type of noise that can really throw off behavioral segments: multiple-user devices sharing an internet access point. For example, a family may all share one laptop. How do you determine if it's a dad using the laptop or a son? The answer is a device or people graph. By creating a uniform master ID from user registration or anonymized PII data, users can be differentiated and tracked across device. Algorithms then look for patterns in content consumption, daypart, or search queries to derive the particular user on a shared device.
Behavioral targeting is also a great way to find consumers in-market for a specific purchase.
Built to support lower funnel goals, the performance of in-market segments will rely on the particular signals used in their creation. Let's say you are a hotel looking to reach people traveling in-market, the best seed data would be to look at people who recently bought airline tickets to your area by using email purchase receipt data. By analyzing the content consumption of these seed users leading up to their purchase, it's possible to predict which other users will be traveling soon.
Recency and frequency are also excellent signals for in-market audiences. Someone who visits car review sites multiple times per week is more likely to be in-market for a car than someone who reads an auto enthusiast site once a month. Because many users purge their browser cookies monthly, or have them blocked by specific browsers altogether, the trick to capturing accurate and consistent reach/frequency data is having registration-level IDs and/or linked mobile device-IDs.
So when someone asks you why there are men in your motherhood audience, check your bias, and ask about what signals are used in the segment. Or watch "Three Men and a Baby." Behavior isn't limited to one gender or another.
Ready to advertise? Contact us today.