Formula 1 pit stop

How to deliver robust MMM models quickly

There are some parallels between efficient and quick MMM delivery and that of a F1 pit stop.

Much like a F1 Pit stop, a lot of things needs to come together for an efficient and quick MMM delivery.

They are:

โ—พ Data Analysis Skills:

One of the biggest roadblock I have seen in getting MMM projects started in my 10 years of experience is data related problems.

Many a time the project gets stalled or deadlines overrun for the sheer fact that some data anomalies were found once the models were built.

It is imperative to do a thorough data check before commencing modeling and report back to the client any data insufficiency or anomalies at the earliest.

A quicker data resolution translates into quicker MMM project delivery.

โ—พStatistical Knowledge

Deep econometric/statistical knowledge is prerequisite to get MMM right. There can be no cutting corners here.

Knowledge of experimentation and causal inference too is required both during model building and post model building for exercises like Incrementality testing, Geo-lifts etc.

โ—พDomain Knowledge

Domain knowledge plays a crucial role in various stages of MMM, starting with Data analysis, Feature selection and Model validation.

One can get many MMM models either through traditional approaches or through open source tools. But the adjudication of whether a model is good or not can only be done in context of the domain.

โ—พProgramming Skills

All the statistical and domain knowledge would be useless if we can’t encode them into the MMM model. Also, good programming skills can make the difference between estimating a set of parameters in seconds to milliseconds.

All these skills prove handy not only to deliver the MMM models faster but also to productize the MMM model. A model that is faster will lead to quicker inference time once productized.

โ—พOpen Source tools:

Open sources tools can also act as a catalyst in quicker MMM delivery. One could also adopt certain functionalities and metrics from these packages to hasten MMM modeling.
For example the automated functionality of carryover and diminishing returns transformation could be taken and added into your own Model.

In summary:

Much like a formula 1 team gets better at pit stop through hours of practice and real experience, Efficient and Quick MMM delivery is also only possible through sheer practice and real experience.

Working much like a F1 pit crew, we have delivered over 350+ MMM models across 8 Industries. And we are only getting better. ๐Ÿ˜Ž

Facebook
Twitter
LinkedIn

Recommended Posts

Chebyshev’s Inequality for Marketing Mix Model Diagnostics

Chebyshev’s Inequality for Marketing…

At Aryma Labs, we constantly endeavor to add as much science as possible…

How to use Robyn’s…

In my last post (ICYMI link in resources), I talked about the similarities…

Similarities between Decomp RSSD and Bayesian Priors in Marketing Mix Modeling (MMM)

Similarities between Decomp RSSD…

Open source Marketing Mix Modeling (MMM) tools are great for democratizing MMM. But…

Scroll to Top