I just read this article (http://bit.ly/12HDmOt) and I guess there are a few things that come to mind as a response:
1) Key concerns:
A) “models underestimate the impact of advertising, particularly of broad-reach network TV; overstate the value of price promotion, mislead marketers into buying thinly rated programming; wrongly downplay risks of going dark for weeks on end; and fail to account for how online search has made all advertising more effective.”
- I’ve always wondered about a model’s ability to determine the short term v. long term impact and this is especially the case with billboards that are difficult to measure impact using statistics.
- I’ve not done any specific research but over-stating the value of price promotions is probably true, because they are so distinct in their impact
- As far as I know high or low rated programming shouldn’t have an effect on MMM accuracy. I could imagine that if you are a major network, you want to keep advertising on your channels and not on the smaller, probably more targeted channels.
- Going dark is always an issue, but if the budget isn’t large enough to be ‘light’ every week, then when money is spent it should be spent optimally.
- On-line search, digital in general and social have special requirements. I think the industry has figured this out.
- Outside validation of a model -I’ve seen some bad models definitely delivering bad recommendations, but this happens in any industry. Most modelers are very good at what they do and deliver fantastic results based on the available data. Maybe there would be some value in developing industry standards. If anyone is interested in something like this, let me know.

2) The question is, can marketers make better decisions without the models? Here I must say the answer is unequivocally NO!
3) Are Marketing Mix Models (based on statistics) valuable? Yes, absolutely.
4) Are there better solutions? Yes (IMHO). For many clients, especially CPG, agent-based modeling & simulation (http://www.ProRelevant.com) is the next frontier in modeling and it is now moving strongly forward.

In all cases, I’d love to hear your feedback.