Such is the case with the notion that you can engage in a media mix modeling initiative, optimize your efforts based on the insights that it uncovers, and feel that for any reasonable length of time it will keep you on course toward safely and efficiently reaching your business objectives. Media mix modeling is typically delivered as a distinct "snapshot in time" project, and not an ongoing, iterative process. It is, in effect, setting your direction by a compass and driving off the road the first time the pavement turns. This method, of course, is not the best scenario for arriving safely to your destination.
The Only Thing That Stays the Same Is Change
Think of the components of your marketing portfolio, some of which include:
· Channels, Campaigns, Tactics, Business Units, Products
· Publishers, Keywords, Websites, Placements, Networks
· Offers, Messaging, Creatives
· Time of Day/Day of Week
· Recency, Frequency
· Target Audience, Segments, Demographics
After looking at this list, you should begin to realize that these components represent only a portion of the factors that influence marketing performance. Your marketing insights and optimization should take into consideration changes that take place over time in the following areas as well:
· Corporate Policy Decisions, Pricing and Products
· World/Local Economic Conditions
· Activities of Competitors
· Availability and Functionality of Publishers, Search Engines, Ad Servers, etc.
Now if any of these factors change after your media mix modeling data is collected, by the time the insights on which you will base your optimization strategies are delivered, they are no longer applicable. A competitor could change its display ad, paid search strategy, creative, bids or spend. A search engine could change its ranking algorithm. The season could change. Your product could have a new feature that drives new creative, messaging and price. Every dimension of your campaigns and the ecosystem in which they exist is constantly in flux, and this flux has an impact on your marketing portfolio's performance. Delivering a single set of insights as a one-time project just doesn't make sense.
Solution = Process, Not Project
This is where the iterative and more frequent nature of the attribution management process provides a more reasonable solution. Once the connections have been made from all your marketing performance data sources, CRM systems, media plans, etc., to an attribution management software solution, that data can flow as frequently as you like -- daily, weekly, monthly. As it flows and an initial attribution model is produced, that process produces insights from which you take actions to optimize your campaigns.
But the advantage that attribution management has is that once your initial set of insights are produced and you optimize your efforts based on those insights, a set of results get generated that flow back into your attribution solution. The attribution model that produced the initial insights then gets validated and course-corrected via machine learning technology. So with each iteration of your marketing results data that is fed into the model, the more and more accurate the model becomes at predicting the results of future changes to your campaigns, resulting in ever-more-accurate insights and ever-more-accurate and effective optimization.
And of course, because changes to your "constantly in flux" marketing ecosystem are being recorded and fed into the model with whatever frequency you chose, their impact can be detected, quantified, and included in the attribution model as quickly as they are recorded. This enables the impact of ecosystem changes to be included when the next set of attribution insights are produced -- which in turn enable you to continually optimize based on the most recent state of your ecosystem.
I prefer to get to my destination without any dents. How about you?
I think your logic may already be dented.
While some modelers still frame their analytics narrowly as you suggest, many of us have moved far beyond and are modeling the full spectrum of variables which you describe above. So the limit of mix modeling isn't the method, it's the modeler. Just like the limit of judgmental attribution isn't the method, it's the quality of the judgments.
Readers of this blog will recognize me as a big supporter of judgmental attribution methods WHEN real data isn't available. But when the data IS available, sampling and judgmental attribution/simulation don't tie back to actual business results nearly as well as econometric methods. And that's the greatest criteria for credibility in the eyes of CEOs and CFOs.
You may be using an early generation GPS. Better have that checked.
Thanks for the comment Pat.
I am not opposed to media mix modeling. But the way it is typically conducted as a one-time project and the way it is typically delivered as a PowerPoint presentation rather than in the form of actual data simply doesn’t meet the needs of today’s data-focused marketers. In that sense, most media mix modeling players in the industry are using the compass and not the GPS. To make the media mix modeling more effective, you need to make it as an ongoing process within the organization. It should take the attribution data into account and there should be interactive output available to those that consume it. The quality of the modeling and judgments are obvious assumptions. Several Fortune 100 companies utilize Visual IQ’s “Top-Down” attribution methodology (one of several attribution methodologies we employ) which is basically actionable and accurate media mix modeling with features and benefits mentioned above.
BTW, I have enjoyed reading your previous posts. They were excellent. You should write more.