MediaWave Actionable Insights and Industry News for Media Professionals
Attribution Models for Campaign Optimization

Campaign Optimization Is Enhanced by Algorithmic Attribution

October 17th, 2016   ||    by Charlene Weisler

Campaign optimization through attribution is a hot topic for digital marketers, 58 percent of whom say it’ll be the top tactic occupying their time and resources this year, up 22 percent from last year, according to eMarketer. But the measurement criteria for attribution is undergoing an evolution from traditional last-touch and first-touch models to more sophisticated computer-generated statistical models that help both buyers and sellers better optimize media spending and consumer targeting.

Algorithmic attribution offers advertisers the ability to react more quickly to competitive changes in the market, eMarketer also noted. But between traditional market-driven forms of attribution and the brave new world of algorithmic attribution, which model allows marketers the flexibility and creativity to better track the path from search to consideration to purchase?

Attention to Incrementality

Whether traditional attribution, like Media Mix Modeling (MMM), or algorithmic, such as Multi-Touch Attribution (MTA), attribution is essentially the practice and measurement of incrementality—the amount of change caused by a small increment of input. In studying the incrementality of attribution, much depends upon the product—cars compared to toothpaste, for example—and the length of consideration time before purchase.

Augustine Fou, Digital Consigliere for Marketing Science Consulting Group, explained at the recent OMMA Programmatic Conference that for some products and services we need to “focus less on the top of the funnel and focus more on the middle of the funnel. Deciding on Lasik surgery takes about three years. So we check search terms. Midfunnel keywords tend to be more precise and closer to the time when the customer will do the surgery.”

But search terms don’t always reflect the consideration process. “Just because you are searching for psoriasis doesn’t mean that you have it,” stated Tyler Pietz, VP Programmatic, Cadreon.

Traditional vs. Algorithmic Attribution

Amy Mitchell, Head of Convertro at AOL Platforms, mapped the differences between MMM and MTA attribution at the recent Advertising Week Conference. She noted MMM understands the soul of the business, taking in all types of touchpoints, from interest rates to the weather to influential communication, but that it also has its difficulties. “MMM is struggling to keep pace because the world is changing,” she admitted, “however, MTA is also struggling in some areas. You can allocate cross-channel change, but it has a hard time including print, sales impact, and economy.”

Ultimately, MMM and MTA are vastly different. As Mitchell concludes, “MMM is a macro, consultative model that has limited consumer data, while MTA is granular and automated, but doesn’t always account for external factors.” For those reasons, you can get different results for the same metric or ad placement.

Automation Advantages

The Coalition for Innovative Media Measurement (CIMM) launched a study with the American Associate of Advertising Agencies’ (the 4A’s) Media Measurement Task Force and Sequent Partners into best practices in cross-platform attribution and ROI analysis. “Algorithmic attribution has potential to be an improvement over rules-based modeling since it allows for more complex fractional attribution across more media. However, it’s still dependent upon the quality of the data input,” stated Jane Clarke, Managing Director and CEO of CIMM.

While traditional attribution might be familiar and comfortable for marketers, algorithmic attribution offers a range of benefits, including:

  • faster response times to market changes
  • intensive analytical applications to changing datasets
  • flexibility with market and business changes
  • fewer opportunities for human error

Further investigation into the most effective forms of campaign optimization with attribution will continue, with the involvement of both traditionalists and data scientists. While the results will vary by product and category, it’s possible a combination of traditional and algorithmic (which could include advanced machine learning) will ultimately prove the most effective and accurate.

Tags: , ,