Commentary

COVID-19 Triggers Innovations In Executing Attribution & Analytic Models

1. Mark Garratt from in4mation insights, November 23, 2020 at 2:57 p.m.

   There has been a lot of talk about how we have to start all over again from March or May 2020. You hear a lot of "past is not prologue" hype. This is not true. There have been tremendous changes, yes, and anyone who was blindly running turnkey industrial MMM is going to be in trouble. The truth is the past is still influential...just less so right now. Two things changed - the baselines and the media impact. Some baseline factors shrunk in importance - like standard macroeconomics or gas prices - others like mobility jumped in importance. Seasonality may have shifted but it still persists. In media, what was already rising up in digital, like CTV/OTT, found new niches and strengthened. Bayesian models have helped us mix the right amount of past data with the current gyrations. Slowly, the past will reassrt itself and will need to be upweighted.

   Reply
Tony Jarvis from Olympic Media Consultancy, November 23, 2020 at 5:27 p.m.
Thanks for the valuable added insights into how key historical data will "return" to something like their former influence within MMM.  Per my report from Day 2 of this ARF/Sequent Partners event, the media data generally being used (notably, so called, "viewable impressions" which has no target audience exposure dimension) fundamentally misleads any model and will consequently provide spurious results concerning any "effects" of the media pressure compared to other more significant "effects" drivers, e.g., creative, trade promotions, pricing, brand equity, etc., etc.!  The headline was, "When Media 'Exposure' Is Not Audience 'Viewing' Or 'Hearing' There Can Be No Ad Outcomes!" 
Perhaps Dr. Jim Spaeth of Sequent Partners will chime in? 
Reply
Ed Papazian from Media Dynamics Inc, November 23, 2020 at 7:22 p.m.
Tony, as you know, you and I plus a few others are on the same wave length on the issue of "impressions" versus actual ad exposure. However, the pressure to do something and do it in a reasonable amount of time underlies this and similar projects and I understand that. But, being fair, the data wont be totally wrong--so there is hope if some adjustment factors are introduced via modelling. As a quess, I would say that approximately 50% of the "ad impressions" whose impact---or presumed impact----will be evaluated are phantoms---meaning that half the time the person counted as exposed wasn't exposed. If this discrepancy could be factored into a parallel analysis of the data, it might reveal that for certain types of considerations---overall audience tonnage, total  reach or demographic targeting, as examples, that it didn't matter as much as it does when the question turns to what's the ideal frequency or how does ad wearout co-relate with impression weight?
Reply

CommentRead more comments >

Next story loading