dannypeled.com
A few years ago, Netflix sought to improve the accuracy and relevance of its content recommendations by offering a million-dollar prize to anyone who could write an algorithm that outperformed Netflix’s existing code. Netflix awarded the prize to a developer team in 2009, but it doesn’t use –and doesn’t plan to use — the winning code.
As the “Prize Master” for this prestigious contest, it’s safe to say that I learned a great deal of information from the Netflix Prize results. But perhaps the most important (and valuable) finding was surprisingly simple: Human patterns are unique, but ultimately also incredibly predictable.
This seemingly small finding has hugely meaningful implications for me today, even as I’ve moved on to solving the problems that face another industry. This revelation has laid the foundation for a recipe that satisfies both marketers and consumers, allowing for something entirely novel — that individuals can actually keep their identities private and marketers can still effectively reach them. Let me explain.
We are all incredibly unique creatures
Something I learned at Netflix that I always found interesting was that if you remove the top most commonly watched movies from a given user’s queue — say, the top blockbusters — it only takes a few of the remaining movies to identify a unique pattern. That is to say, a relatively small amount of unpopular movies from a random user’s queue can be used to reasonably pinpoint that specific user with high accuracy at an anonymous level. If we take this principle outside of the Netflix realm, this means that just a tiny fraction of our overall digital activity is enough to make predictions about us at an anonymous level. That’s how unique we are as individuals.
However, we are also creatures of habit
Considering how unique we all are as individual consumers, it may seem counterintuitive that we are therefore predictable, as well. But uniqueness and predictability are not mutually exclusive. For example, while my behavior as a consumer, employee, father, etc., are very different from yours, ultimately they lead to recognizable patterns, and human patterns are actually very predictable. For Netflix, this means making predictions about the movies we watch.
In the digital marketing and advertising industry, this same concept is applied to understand consumers as they work, socialize, research, and purchase products in today’s multi-device world. We’re all unique — I own a certain set of devices and visit certain websites and use certain apps, which all makes me relatively unique — but also predictable. There are patterns to when I do things on each devices — news in the morning on my smartphone, work-related sites during the day from my laptop, games at night on my tablet, for example — and savvy marketers can use those patterns to provide relevant marketing offers.
Accuracy does not require exchanging your personal identity
Everything I’ve described above is possible at an anonymous level. Today there are several companies that can make predictions about what devices I use and what sites I visit. That information helps those companies make decisions about how to customize my digital experience, whether that’s personalized landing pages or targeted ads and deals.
Knowing my name and birthdate and other personal information isn’t necessary for marketers to deliver a strong user experience for consumers, just like Netflix doesn’t need that to determine which movie I’m likely to watch.
All that’s needed is patterns.
Today, brands are enabled to customize experiences for consumers based on these online activity patterns, just like the Netflix algorithm does. But while some advertisers use a predictive model, other companies such as Facebook and Google, provide a way for marketers to use consumers’ personal data for reaching audiences.
It doesn’t have to be that way. The predictive, or probabilistic, model is the real prize. This is the best way for marketers to deliver custom experiences, and for consumers to retain their personal information. A deterministic approach, in addition to having privacy concerns, doesn’t seem necessary, given everything that predictive algorithms can do.
So, what do you think? Do probabilistic, predictive algorithms seem like the best way to reach audiences?
Devin Guan is the vice president of engineering at Drawbridge, the leading programmatic cross-device technology company, where he brings more than 15 years of experience as a leader, entrepreneur and developer. Prior to Drawbridge, he was chief architect at Announce Media, where he built a technology platform that powered the business to secure $260 million in annual revenue. Guan joined Announce Media from Netflix, where he was responsible for conducting the Netflix Prize, a large-scale crowd-sourcing machine learning contest. He was also as a founding member of AdMob’s optimization team (acquired by Google in 2009), and has held various engineering leadership roles at prominent companies, including Yahoo. Reach him @gnived.