Sramana Mitra: Talk to us a little bit about all these different data sources. What are the strengths and weaknesses? What level of depth can you find?
Diaz Nesamoney: The differences and, arguably, the challenges with data is two-fold. One is some data is essentially not very accurate because it’s based on projections or modeling. They call it look-alike modeling. There, the data is often wrong. That’s one problem. It turns out that the data we collect from the brand is very precise. We tend to lean more on that than all the other data. The other data is used to supplement.
The second is completeness of data. We might know three things about you and we might know three entirely different things about another person but we don’t know the six things that we need to know to effectively message them. That’s challenging. For both of these problems, we are applying machine learning to fill the gaps. With sufficiently large amounts of data, we can pattern match and say, “This person has certain attributes that are similar to this other person.” This is another thing that Amazon has been doing for years.
Sramana Mitra: Collaborative filtering.
Diaz Nesamoney: It’s used quite a bit in the world of e-commerce. We’re borrowing that concept into the world of digital advertising where completeness of data is always going to be a challenge. We’ll never know everything there is to know.
Sramana Mitra: What are you learning from ROI models and so forth of what kind of impact your personalized algorithms are having?
Diaz Nesamoney: That’s the great part of this. In spite of being a big believer in personalization, I’m blown away by the results that personalization is delivering. There’s a funny anecdote. Last year, we were working on polishing a benchmark. I told our team to pull up the data. We pulled up a billion ad impressions across 24 different brands globally.
The team came back and told me that, on average, we were seeing 3x of performance against a non-personalized ad. I said, “Come on guys. This sounds a little too optimistic. Do it again.” They did it again. They came back with a few adjustments, but it was pretty close. I said, “We’re going to publish this. Our credibility is on the line.”
Our Head of Products was sitting in my office. I asked, “Are you sure before we publish this?” She said yes. I said, “Do me a favor. Log me into one of our client’s accounts. I want to see it.” I picked a random one. I looked at the number and it was there. That’s what’s exciting about this. It does deliver anywhere from doubling the engagement to tripling the engagement, which is phenomenal in the world of advertising. If you step back and think about why. It’s pretty obvious. Consumers are just sick and tired of irrelevant ads. They don’t mind ads. We’re just being bombarded with ads that are completely irrelevant.
Sramana Mitra: That’s the beauty of Facebook.
Diaz Nesamoney: Exactly. Why show somebody an ad that they will never be interested in simply because of their gender, demographic, or where they live. We’re seeing that. Our clients are seeing that. Almost everyone of the clients that we signed a couple of years ago are still with us.