Marketing Mix Modeling (MMM) – Optimize your Marketing ROI

How understanding seasonality and cyclicity can help you build better MMM models

Seasonality ≠ Cyclicity Be it time series analysis or Marketing Mix Models (MMM), the distinction between seasonality and cyclicity is important. Many confuse seasonality with cyclical time series. Here is a quick distinction. Seasonal: A seasonal pattern is a fluctuation which occurs at regular time intervals. These time intervals are predictable. Seasonality does not always refer to changing seasons. It could be anything which happens at predictable and regular intervals. The frequency of these events is fixed and is associated with calendar events. E.g., electricity demand pattern or sales trend showing surges around holiday season. Cyclical: A cyclical pattern is a time series which does not occur at regular intervals.

Standardization before Regularization in MMM

Standardization before Regularization in MMM In my last post (link under resources), I covered the topics of Multicollinearity and Endogeneity. And how solving for Multicollinearity can lead to Endogeneity. To solve for Multicollinearity, many adopt regularization like Lasso or Ridge. But here are some key points to keep in mind. 👉 L1 and L2 are the most common regularization techniques. However, one common mistake while using this approach is not applying standardization to the data. 👉 Many libraries do not explicitly tell the user to ‘standardize’ their data before applying regularization by default! 👉 L1 and L2 techniques penalize large coefficient values more while applying shrinkage. If the data are not

How common are S-curves in Marketing Mix Models (MMM)?

How common are S-curves in Marketing Mix Models (MMM)? Before we dwell on the S-curves, let’s talk about the Hill function. Did you know that the Hill function used for transforming media variables to capture diminishing effect has its origins in Biochemistry and pharmacology!! The equation was formulated by Archibald Hill in 1910 to describe oxygen binding to Hemoglobin. Hill Function Formula w.r.t Marketing Mix Model (MMM) is in Fig 1, where: S, the shape parameter, or slope, decides what shape the response curve will take: S curve or C-curve. If S>1, the curve follows an S-shape If S <1, the curve follows c-shape Notice in Fig 2 that at

Multicollinearity is one of the enemies of Marketing Mix Modeling (MMM).

Multicollinearity is one of the enemies of Marketing Mix Modeling (MMM). Multicollinearity generally occurs when there is a high correlation between independent variables. Multicollinearity does not affect predictive power of the model but it causes a lot of issues when one needs to infer or attribute the changes in dependent variable to the independent variables. In the context of Market Mix Modeling or other attribution models, Multicollinearity can be a big problem. One solution that many adopt to solve multicollinearity is – ‘Drop the variable’. But solving multicollinearity through this method leads to another problem – The Omitted Variable Bias (OVB). OVB in turn leads to Endogeneity. Endogeneity happens when

CMMMO = CSO

The Chief Scrutinizing Officer (CSO) Every Marketing Mix Modeling (MMM) project or for that matter any Data Science project needs a Chief Scrutinizing Officer (CSO). Here is how the skill of scrutinizing comes in handy: 1. Scrutinizing the data: The first step in any Data Science is to examine the data carefully. In our recent projects, we found that the media investments for a particular medium were in a range of 100k, but then one data point showed value in millions. We requested the client to recheck the data and it turned out there was indeed a mistake in capturing the data!! 2. Scrutinizing the models: This phase involves asking the

MMM Model Update – What, Why and When

Wondering when to update your MMM model? Here is a guide 👇 Predominantly, a Machine Learning model needs to be updated for two reasons: 1. Model Drift 2. Business Reasons Let’s first breakdown scenarios where MMM models need updating. ◾️ Seasonality – Some brands in CPG/ FMCG space exhibit seasonality. For e.g., Packaged Juices may sell more during summer. If your models’ last few data points were say of ‘Winter months’ then the same level of sales demand may not be applicable to the spring and summer months. To factor this, MMM model needs to be updated. ◾️ Campaign Launch – Sometimes post model building, the brand might have launched

What is the best Marketing Budget Allocation

Marketing Budget Allocation One of the important use cases of MMM is the Marketing Budget Allocation. The results of MMM are directly leveraged to provide various ‘What-if’ budget allocation scenarios. Such as: ▪️ Under the same budget, how could we have allocated the spends across media differently to yield a lift in KPI? ▪️ What should be the allocation across media if there is a 30% increase in budget? ▪️ What if we decide to cap media A’s allocation to 10% of budget, what would be the net effect on KPI? While the clients are normally happy to see these results, one question that most clients as well as prospective

Marketing Mix Modeling (MMM) and MTA are like X-ray into your marketing strategy.

Marketing Mix Modeling (MMM) and MTA are like X-ray into your marketing strategy. Much like an X-ray. They tell you: ✅ Which strategy is broken. ✅ Is your revised strategy healing (performing) well. ✅ Does any anomaly point to future problems. Given the recessionary trends worldwide, now more than ever MMM & MTA are the need of the hour for any organization. I have had interesting chats with CMOs, CROs, Brand Managers, Performance Marketing Managers and Media Planners of late. This is what I glean from the conversations: 💡 Being prudent with marketing spends. Organizations want to be more prudent with their marketing spends in 2023. They want to maintain

7 key steps to get your Multi Touch Attribution (MTA) right

7 key steps to get your Multi Touch Attribution (MTA) right !! We have built 30 Markov Attribution models for companies across geographies in the last 3 yrs. Here are the 7 key steps to make your MTA project a success. 📌 Reality is different from toy examples: Things don’t work as easily as illustrated on toy examples or tutorials. Markov Attributions implementation on a real data set is a different beast altogether. This expectation needs to be set. 📌 It is all about the Paths: Roughly, 60% of the Markov attribution projects is about Data pre-processing and data transformation (forming the paths). It is pertinent to get this step

Don’t Stepwise Regression your MMM model

So recently a client hired us to build MMM models for them after failed attempts to in-house the MMM capability. Earlier their in-house Machine Learning engineers (with no statistics background) had built their MMM models thinking that it is just ‘linear regression’. We sat down with the MLEs and wanted to know how they went about building the models. The MLEs said they chose variables based on ‘Trial and Error’. Which then turned out to be backward stepwise regression. If you want a refresher on what is stepwise regression (links are in comments). Further we noticed that they had not applied proper adstock on any of their media variables (but

Blogs