So Google just deprecated third party cookies for 1% of users worldwide. In nearly 270 days, third party cookies will be totally deprecated.
The domino effect of this would be that Marketing Attribution will get more tougher.
But Marketing Mix Modeling (MMM) is a good solution to fill the void.
We are now seeing a lot of companies trying their hand at MMM.
One fundamental question that many first time adopters of MMM have is – Should we try Bayesian MMM or Frequentist MMM ?
We would suggest go with Frequentist MMM and here is why:
📌 Prior Elicitation:
In Bayesian MMM, the vendors make the client work. The vendors would ask you “hey what do you think the ROI of youtube ad should be”, “What do you think ROI of TV should be.”
As a first time adopter of MMM, you will have no clue on this. But the funny part is, the Bayesian MMM vendor too won’t have any clue 😅 and will plaster some uninformative priors.
This inevitably leads to a situation where your model is biased from the start and you will end up with inaccurate models.
I know some supporters of Bayesian MMM would say “With more data, prior choice will hardly matter”. Yes but MMM is a small data problem (relatively speaking).
If your priors are way off, it will simply overwhelm the evidence in your data.
📌 Complexity and Time:
While Bayesian MMM vendors may sell you the beautiful philosophy “we update our beliefs as we see new evidence”, the reality of application of Bayesian methods is far from rosy.
Bayesian MMM takes more compute time as inherently Bayesian methods don’t offer a closed form solution. You will have to sample the posterior distribution and use memory intensive MCMC methods.
Bayesian MMM takes longer to run and makes the whole process unnecessarily complex.
📌 Cost:
Because of all the unnecessary work that is involved in Bayesian MMM, Bayesian MMM is costlier than Frequentist MMM. All that extra compute that happens in Bayesian MMM, accrues a cost.
That cost is inevitable passed on to the end client (you).
📌 Statistical Reasons – Multicollinearity
Multicollinearity problems always will persist in MMM. But Bayesian MMM is very fragile when it comes to even moderate multicollinearity. See my detailed post in resources.
🎯 In summary –
Frequentist MMM offers a comprehensive solution sans the complexity. Bayesian MMM offers no guard rails and there are many ways to go wrong.
Also I will let you in on a industry secret. Bayesian MMM vendors also *discreetly* build a Frequentist MMM alongside Bayesian MMM, just to see if they are headed in the right direction. 😅
In a way Frequentist MMM is used as a benchmark.
As a first time adopter you want quick and accurate results and not complexity, inaccurate results and inflated bills.
Reach out to us if you want to adopt MMM without breaking your bank and don’t want to be mired in complexities.
Resources:
Bayesian MMM is not a silver bullet for MMM’s Multicollinearity issue :
https://www.linkedin.com/posts/venkat-raman-analytics_marketingmixmodeling-marketingattribution-activity-7106973443807444992-H5Bq?utm_source=share&utm_medium=member_desktop
The Richard McElreath’s Quartet and problem with priors :
https://www.linkedin.com/posts/venkat-raman-analytics_marketingmixmodeling-marketingattribution-activity-7107946791983071232-w07F?utm_source=share&utm_medium=member_desktop
