Day One of the Advertising Research Foundation’s (ARF) sixth annual “Attribution & Analytics Accelerator” conference, in partnership with Sequent Partners, raised as
many questions on attribution modeling approaches and techniques as it answered.
Ross Link, CEO of Marketing Attribution LLC and Jeff Doud, director of advanced analytics and
reporting at Ocean Spray Cranberries Inc, kicked off with an intriguing title, “Measure Small, Measure Through Walls with RCTs (Randomized Control Trials).” It underlined the
fundamental concerns with any modeling approach that attempts to measure the effects of the full ad ecosystem with all its complexities.
These concerns include,
“walled data gardens” -- notably Google, Facebook, and Amazon -- the lack of actual persons data due to required privacy compliance (households have become a surrogate), test group
selection bias, and incongruent cross-media audience data.
Presenters suggested that using an “intent to treat” RCT version of multi-touch-attribution
modeling, MTA, Ocean Spray has been able to measure channels that MTA could not, along with the added value of real time optimization and resulting investment improvements.
This approach applies a persons-based analytic approach to test groups from massive 10 million household samples with daily sales and media impressions data.
Doud
underlined that, while the test versus control matches were good, the measured effects were small when slicing data this thin. He reminded us that advertising, while powerful, is overall a weak
force. Another measurement/modeling concern to be addressed?
Justin Toman, head of sports marketing at PepsiCo and Adam Holt, senior vice president sales/partnerships at
FanAI, reviewed how the emergence of sponsorship attribution is allowing PepsiCo’s sports marketing team make better decisions about the level of investment appropriate for major sports teams
they partner with. This presentation raised the question of whether an attribution modeling approach could also be used by major arts groups?
Yes, modeling can help with this
amorphous marketing opportunity. Their inclusion of the value of long-term fan equity associated with a sports team as part of the input to assess the effective target level investment seems
spot on.
Will the Metropolitan Opera be next?
Lana Koretsky, director of TV & online video marketing at Wayfair, reminded that
measurement and modeling is an evolving process. It is therefore important to focus on progress rather than perfection when examining campaign outcomes, despite ongoing data quality
issues.
Wayfair moved its entire TV/video operation and MTA modeling in-house several years ago. It relies on iSpot.tv impressions data to help it understand the
effects of linear and digital TV in combination, based on both short- and longer-term value to the brand.
In recognition of the relative superior power of the creative
message versus media’s contribution, the use of on-the-fly multiple creative executions and evaluating the resulting changes was noteworthy.
Can marketing mix
modeling (MMM) be used to include augmented reality (AR) and its effects and contributions to fast-moving consumer goods’ offline sales?
Takeshi Tawarada, a
researcher at Snap, Inc., concluded there is evidence AR strengthens connections to the brand, driving engagement and attention. He referenced AR impressions as part of the MMM input.
Whether these impressions are comparable or equivalent to impression metrics used for other media platforms remained an open question.
According to Jim Spaeth, a partner
at Sequent Partners, attribution and campaign modeling approaches continue to make improvements. As noted last year, perhaps these models are still doing precise things with imprecise -- or
misunderstood or misrepresented -- data? The overarching issues of data availability, data quality and effects of privacy requirements on persons information remain.
Is the goal of attempting to give credit -- across all media touch points -- paramount to advertisers and achievable at any meaningful level of accuracy? If “impressions”
used across each media platforms in any model, are of different basis, just how accurate can these models be?
Stay tuned for my commentary on Day 2 and further insights
on the art of attribution modeling versus the science.