Ep. 4 - Marketing Mix Modeling: How to Transform media with Diminishing Returns and Adstock on Excel
ฝัง
- เผยแพร่เมื่อ 17 ก.ค. 2024
- A robust model should make sense both statistically and commercially.
Raw variables need to be transformed to reflect true consumer behavior.
Check out the most popular transformations in Marketing Mix Modeling (MMM), such as Diminishing Returns and AdStock.
Diminishing returns measure the saturation point of each channel
Adstock effect measure what's the advertising investment's impact over time
If you apply these transformations using Excel and you'll create a more accurate model business wise and statistically
Video structure:
0:00 Intro
1:45 AdStock & Diminishing Returns Formula
3:45 Transforming in Diminishing Returns & AdStock
6:55 Model hyperparameters optimization
11:00 Generate predictions and test accuracy
12:00 Show Model's accuracy graphically
14:00 Error analysis
15:50 End Notes
TEMPLATE: bit.ly/excelmmm
If you want to reach out:
Marketing Mix Modeling Service: bit.ly/3vcopZA
Linkedin: bit.ly/3p9rkON - วิทยาศาสตร์และเทคโนโลยี
best mmm video so far!! absolutely love love it! thank youuu for making this video
I am a data analyst, working in the media space in Nigeria. Your videos are great.
Thanks Maxon!
wow this is interesting! thank you
Thanks so much for putting together these videos! In this example is the Intercept the same thing as baseline sales? In other words if We had no advertising in the measured media we would expect negative sales?
Hi Michael, you are right, the intercept equals the baseline.
In the diminishing returns case we happen to receive a negative value on the intercept.
For a complete MMM project you don't want that.
It's always better to obtain a lower R^2 but with coefficients that make sense in real life.
Thanks for the simplest way of explanation. I have one question for you. How do you interpret the Alpha values in business terms? Beta explains the carryover/Decay effect but Alpha I couldn't interpret. Please help.
alpha is used to describe diminishing returns
the lower the alpha is and the lower is your saturation point
The formula we use is the following:
Response = Spend^(Alpha)
Alpha value between 0 and 1
Hey! I need to use adstock and diminishing effects for my thesis. Which literature would you suggest me to use to base this formula?
Hei Ibra, what do you mean specifically for "literature"?
also what is the reason for using the LINEST function to get the RMSE if the relationship between the independent and dependent variable is non linear?
The relation between the raw indipendent variable and the dependent is not linear, but the relationship between the transformed indipendent variable and the dependent is linear.
At least we need to find a linear relationship.
Also we use linear regression for the most part because it's incredibly easy to analyze and derive marketing experiments off of it.
To give you prospective, in our workflow we also use Neural networks, but they are incredibly more complicated to analyze and for most projects, even tho they are super precise, they are not useful
I don't have REGR.LIN in my excel. I am unable to find how to add this to excel. Can you help?
Hello, that's just the keyword for "Linear Regression" in Italian (our native language). Based on the language of your excel that might vary a bit, you can google that up and find your exact formula for that.
Thanks for this- what DR function do you use?
We use a power function on excel
Coeff* Expentiture ^ alpha
Alpha is a value between 0 and 1
@@cassandra4533 any chance you can please demonstrate the exponential DR function?
What can I do if even after optimization, my R squared values are weak... like 0.125?
Hello, in that case we'd suggest to go back and start from the input data.
An RSQ that low could be caused by a number of things such as:
1) Incorrect input data
2) Incomplete input data
3) Completely missing relevant input variables that affect your output variable
4) Not enough historical data, if it's a new business with low spend and revenue volumes you might simply not be able to use MMM yet
@@cassandra4533 Yes, that's what is happening in my case. Thank you!
Hi! Does someone knows about lightweight MMM? If so, I would like to get the predicted values from plot_model_fit, how can I get those? I've tried using he predict method also doing the inverse_transform over the values but the output is really strange (multiple negative sales amounts which make no sense).
Hello Tomas, we would suggest asking directly in Lightweight's GitHub.
@@cassandra4533 thanks for your response. Aslo thanks for all the MMM videos, they've really helped me grasp the logic behind it!
Do we need to check for statistical significance? Using p-value or something?
Yess for any regression analysis is always best practice to validate the coefficient measuring the p-value.
We do that analysis in the python MMM
@@cassandra4533 Is there some way to also check it in Excel, especially when number of independent variables is on the higher side?
what is the reason for transforming the data?
Hi Benjamin, thank you for your comment.
Marketing wise we need to consider the adstock phenomenon (not all the orders happen the same day of the investment) and diminishing returns (incremental media response gets lower with higher investments)
We transform the data to create a model that simulates reality correctly.
so in short.. your Alpha is your coefficient value .. and that beta is your retention rate ! let me know if i sound confusing
Not exactly right mate: Alpha is the diminishing return rate, beta is retention rate, then you have the coefficient from the regression
@@cassandra4533 suppose i have 250 support point where i want to see my variable curve .. through your method it wont work right!..
4:59 Adstock transformation