Don't Hate Triangulation, Embrace it

MMM + Experimentations + Causal Inference = True MROI

There is a lot of talk about Triangulation in the MMM space. Many vendors have some variant of this triangulation feature.

However some hate it !!

📐 But what is triangulation ?

Triangulation is used in different domains - Navigation, Surveying, telecommunication and Scientific Research.

One overarching definition is that, it is a process through which one could hone in on a (object/place/or some phenomenon) through multiple points of references. If the number of reference points are three, then it is termed
as triangulation.

🎯 Why Triangulation makes sense in MMM and Marketing Measurement in general.

Triangulation in MMM makes sense because there is one true Marketing Mix Model .

In MMM we deal with historical data, meaning the sales (or any KPI) has already been realized.

Now the million dollar (literally) question is what lead to this sales? What combination of factors led to the sales.

Now because this is all in the past, there is only one set of combination which could have resulted in the sales.

The job of a MMM vendor is to find out this combination of factors that led to the observed sales. So technically it is honing on the truth.

There are no multiple ways through which sales could have been gotten.

Now that we have established that there is one true MMM model that led to the sales that we observed, we need to devise a mechanism to find the true marketing effectiveness numbers of these factors.

🔺MMM is one part of the Triangulation

One way to find out the true marketing effectiveness numbers of marketing channels is through MMM.

But MMM alone is not enough. We need a way to verify whether the numbers we have got are correct or not. Now there are multiple ways to validate this:

📊 Perform Experimentation

Experimentation can help provide directional idea of the ROI numbers (it can only be directional because experiments are short term in comparison to MMMs).

🔁 Causal Inference

The third option is to perform causal experiments to causally prove the efficacy of MMMs.

Our Triangulation hence looks like the below:

❓ Why some hate triangulation?

People hate triangulation because they misunderstand it or apply it wrongly.

One should not use granular campaign level models/experiments to validate cross channel comparisons or vice versa. One should ideally use another granular level model to validate granular models/experiments or cross channel (meta level) models to validate other cross channel models. In a way it is like a fractal of triangulation .

We in our recent case study demonstrate how to validate the in-platform ROI numbers with a sub granular level MMM.

We could not only tell which campaigns (ASC or Pmax campaigns) drove sales the most but also could tally up the sales to marketing numbers with the bigger MMM model (see case study for details link in comments).

In summary: Don't hate triangulation, embrace it :)

---

Thanks for reading.

Need help with MMM implementation? Click here

Stay tuned for more articles on MMM.

---