Interview: Targeting is Becoming a Data Science Problem

With the impending loss of the cookie and changing nature of tracking user IDs, what does the future of targeting look like? Moving forward, advertisers must reference and merge a plethora of user attributes to make predictions about target audiences. As such, the traditional boundaries of targeting will begin to erode — becoming more of a data science problem. The interview below with Miguel Araujo, Director of Data Science at Semasio, investigates these ideas.

insideAI News: With the impending loss of cookies and continuing evolution of tracking consumers, how has targeting become a data science problem?

Miguel Araujo: Widespread consumer tracking started a new paradigm for advertisers, where fine-grained data about virtually everyone’s online history enabled rich behavioral targeting. It was relatively simple to follow a user across websites — bombarding consumers with products left in abandoned shopping carts. It was also common to define targeting audiences using taxonomies layered on top of which websites one visited, so an advertiser could target consumers who, for instance, visited traveling websites in the past couple of weeks.

As that will be going away for a significant percentage of the internet population, advertisers will be looking for ways to simultaneously maintain their targeting accuracy and flexibility. This is an area ripe with interesting data science challenges given its scale and importance. There are opportunities to leverage contextual information (using the content of the page being visited) — but also to develop new techniques that deal with a fragmented ecosystem where different operating systems, browsers, websites, publishers and third parties all send different sets of information to our platform. With partial and missing data on each request, we will see a unification of targeting approaches as the line between behavioral and contextual techniques blurs.

insideAI News: How are you working to solve this problem as the new director of data science at Semasio?

Miguel Araujo: Contextual targeting will become ever more important, as there are obvious benefits in showing your ad on the right web pages at the right time. But there are amazing opportunities for those able to understand not only the content but also the tone and mood of a publication. This is also extremely relevant for advertisers, as good ad placement is part of your brand image and one needs to know whether page-brand fit exists.

However, let’s not think that behavioral targeting is a dead-man walking. The loss of third-party cookies is not the end for consumer identifiers, and there are many initiatives around combining multiple IDs and entity resolution bubbling underneath the surface.

insideAI News: What do key players in the data science space need to know about this trend?

Miguel Araujo: Data science experts will see the similarities between this shift from behavioral to content in the ad space and what happened to recommender systems some years ago. While collaborative filtering techniques (i.e., basing recommendations on what other users who like the same items I do also liked) brought huge improvements in recommendation quality, subsequent years saw increased efforts on understanding context to solve the cold start problem. What do we recommend to someone who is visiting our website for the first time? The fact that the consumer opened a link to a red cocktail dress becomes so much more important, and so does having an in-depth understanding of all our inventory. While the amount of data in online advertising requires a whole different scale, the good news is that we have seen this transition in other industries before. We have the right tools.

insideAI News: Moving forward, how can advertisers and data scientists work together to overcome this challenge?

Miguel Araujo: Understanding the nuances of the industry requires significant collaboration. Consider how different it is to work with media agencies and with big brands advertising directly. Think about how different products also require naturally different strategies. For instance, trying to sell a product on a heavily saturated and undifferentiated market (such as telecommunications) has a focus on churn that electric car brands do not require at this point.

However, I don’t think that will be the biggest collaboration front. There are two big parts to finding the right target audience: describing who your target consumers are, and understanding who they are and how they behave. More than having data scientists and advertisers working together, data scientists need to create a future where advertisers and software work together. This collaboration will require sharp data-driven collaborative tools, new interfaces and layered user experiences — bringing data insights to the forefront, and so much more because at the end of the day, targeting and understanding behavior are just two sides of the same coin.

About the Interviewee

Miguel Araujo is Director of Data Science at Semasio, where he’s responsible for leading the data science department from R&D to finished product. Araujo joins the company from Feedzai, where he most recently drove omnichannel and data sharing strategy as Data Science Manager. Holding a MSc in Informatics from the University of Porto and a PhD in Computer Science from Carnegie Mellon University, Araujo was also an invited Assistant Professor in the Masters of Data Science at the University of Porto. 

Sign up for the free insideAI News newsletter.

Join us on Twitter: @InsideBigData1 – https://twitter.com/InsideBigData1