AttributionData Science

Beyond Black Box Attribution, Part 2: Exploring the Ensemble Method

By January 26, 2016 No Comments

I’m calling it: Black box attribution is officially on its way out. It’s no longer necessary to blindly trust that your attribution platform will magically turn data into meaningful insights. It’s time for the data, and the insights, to escape the black box and join their technology friends in the light.

To that end, let’s take a closer look at Conversion Logic’s methodology: the ensemble method.

As discussed in Part 1, the ensemble method is an approach to attribution that uses multiple algorithms instead of just one. While single algorithms each excel at certain applications, whether answering a specific business question or measuring marketing efficiency, they also have pros and cons. That means that there is no single model that covers all the bases, and banking on a single-model approach can be inaccurate and limiting.

The ensemble framework blends individual models’ results to achieve a prediction accuracy that is higher than any single model in the ensemble. In layman’s terms, the whole is greater than the sum of its parts. Using machine learning, the resultant model combined from multiple algorithms is also individually tailored for each clients’ data set, rather than a one-size-fits-all approach.

The same concept was explored recently in “The Wisdom of Crowds,” by New Yorker columnist James Surowiecki. He came to a similar conclusion: multiple predictions from a “crowd” (whether of people or statistical models) will generally be more accurate than any single prediction on its own.

Surowiecki provides an interesting real-world example in the book. In an experiment that originally took place at a county fair in England, NPR’s Planet Money posted a picture of a cow online and asked people to guess its weight. At 17,000 guesses, the average weight guess was within 5% of the original weight. Surowiecki says, “There’s something magical about it. It’s not magic. It’s just math, but it seems magical.”

Which brings us back to the black box. Our ensemble method isn’t magic, either, but the math is compelling. In our recent article on CMO.com, The Wisdom of (Algorithmic) Crowds: Why Single-Model Attribution Isn’t Enough, I broke down the potential savings that a marketer could achieve by increasing predictive accuracy. The ensemble approach can improve results by up to 35%, which translates to serious money:

  • $100 million budget: $25-$35 million potential savings
  • $1 billion budget: $250-$300 million potential savings
  • $600 billion, the entire global advertising industry: $15-$21 billion in cost savings

Magic has its time and place, but attribution isn’t one of them. The technology, market, and business models have all evolved so that we not only can share what’s happening in the black box, we should. A clear understanding of how the attribution platform works means more visibility, higher adoption, and deeper insights for marketers – all of which lead to greater success.

Leave a Reply