What Happens When Modelers Challenge Each Other?

  • by , Op-Ed Contributor, November 16, 2022

Day Two of the Advertising Research Foundation’s and Sequent Partners' 2022 “Attribution & Analytics Accelerator” conference featured a challenge from one of the industry’s most respected researchers to another.

NCSolutions Chief Research Officer Leslie Wood challenged Marketscience Executive Partner Peter Cain which of its two attribution models was most relevant and meaningful to marketing decision-makers.

Both approaches clearly have their merits, and both are using either machine learning or AI.  However, when two modeling super stars collide clients ultimately win.

Tuesday’s sessions were themed “Tales from the Far-End of the Cutting Edge: AI & Beyond,” but they could well have been subtitled, “Can it be Trusted?”  This is “hairy” media/marketing research and modeling stuff, but the sessions did not disappoint.  

Nielsen Senior Vice President-Product Management Katie Korval focused on data challenges for modelers in her presentation, “Maintaining Confidence in Attribution Amid Industry Changes.” She underlined the issues – digital disruption, identifying audiences across channels, devices, and ecosystems in a non-cookie world – along with the massive fragmentation of markets and media, as well as consumer privacy.  



Korval also noted the critical importance of data harmonization and interoperability, data security, confidentiality, and control, as well as validating the accuracy and representativeness of it. Reminder: Nielsen still operates a massive panel that serves as high quality reference across-the-board.  Nielsen also has been increasing its collaboration with other data providers, including those with walled gardens.  

Marketscience’s Cain stressed the value of going beyond conventional marketing mix models and using econometric principles to produce a dynamic time series model that would address both short- and long-term causality for more comprehensive insights.  This approach, he said, “eliminates the last touch bias” of conventional marketing mix models. 

A believe a presentation delivered byAudi Switzerland Team Leader-Digital Marketing Filip Pujic was truly meaningful. Audi uses “attention units” sold by Adelaide as part of its media planning and buying.  Pujic demonstrated the value of integrating this persons attention-based metric, which significantly improves programmatic media bidding, which I think is an important lesson for the U.S. media scene?  

After more than eight years of intense testing and analysis, NCSolutions’ Wood, believes it has mastered the promise of machine learning for measuring incremental sales with its “NextGen” model. The model’s unique attribute is that it can estimate “counter factual insights” to provide sales lift for non-exposed as well as ad-exposed household.  One major finding: the bigger the brand the more prospect targets matter to media execution. 

In view of the very different currencies use by media platforms, I wonder how so many media metric/ad exposure inputs for any campaign can be equivalized.  Perhaps Audi has the optimal solution.

“Producing any analytics data set is extremely complex and excruciatingly difficult,” concluded Plus Company Global Chief Data & Analytics Officer Michael A. Cohen, adding, “AI is certainly having an impact on attribution models and making marketing decisions.”

To me, it seems trust is being built.

Next story loading loading..