Modern Marketing Mix Modeling Explained

Guide to Modern Marketing Mix Modeling.

Proposed by the top modern MMM software: https://cassandra.app

You are going to learn:

- Why you should adopt MMM
- Data required for MMM
- Common MMM miths
- How much should you invest to start adopting MMM
- Procedure of adopting MMM
- Triangulating Measurements
- Strategic MMM applications

1. Modern
Marketing mix
Modeling Explained

2. Intro
02
•Myths around
MMM
•Strategic
applications
•Why MMM
what we’re going to cover:

3. Application
18
All Marketing
Operation
and spend in
one place
Increased
confidence
in Marketing
allocation
decisions
Generate
more returns
from the
same budget.
Plan & Forecast
Marketing
Investments
more
accurately

4. Intro
03
What is Marketing
Mix Modeling?
Revenue
Invested Budget

5. Intro
04
How it works
Media
Organic
Context
variables
Overall Sales
Offline and Online
ROI measurements
Incrementality
analysis

Seasonality
analysis

Media
Saturation
Estimates
Media Mix
Allocator
Marketing Mix Modeling transforms your historical Marketing
data into strategic measurement and optimization insights

6. Intro
05
Digital tracking like GA is deprecated due to
privacy regulations and cookies sunset.

Scrutiny on how efficiently brands allocate their
marketing investments.

Measuring Offline and Online Media ROI and
optimize them.
Gold era of digital
measurement
GDPR & Privacy
regulations
Amazon and Meta
release of MMM
programs
iOS 14 release &
release of Google
MMM programs
Privacy focused
measurements

7. Myths
06
Myths
MMM is a blackbox
Marketing Mix Modeling is useful
only for companies that invest tens
of Millions every year
MMM can only be used
once per year
MMM is extremely expensive
MMM lacks of real
time feature
06

8. Case studies
07
MMM is NOT a blackbox
Marketing Mix Modeling uses a fairly easy
regression type analyses combined with non
linear transformation of its data.
Robyn, one of the most used open source library
uses Ridge regression and transformations like
Weibull Adstock and Hill Diminishing returns.
You can find the logic in their repository with no cost.
Myths

9. Case studies
08
Myths
Marketing Mix
Modeling is useful if
you spend at least
30k€/month
Marketing Mix Modeling is an analysis
that solves complexity. The
more complexity you have in your
media mix the more a Marketing mix
modeling can help you
10k
€/month
20k
€/month
100k
€/month30k
€/month
1
Channel
2
Channels
5+
Channels
3-4
Channels
Advertising Spend vs MMM Complexity/Validation
LOW
MODERATE
HIGH
VERY HIGH
After reaching 30k€/month
of advertising investments
your volumes are sufficient to
validate the results and your
investments start to become
complex enough to require a
MMM

10. Case studies
09
Modern Marketing Mix Modeling solutions like
Cassandra allow you to refresh your
MMM monthly allowing it to improve in
accuracy and robustness. 
MMM can be
refreshed monthly
Thanks to this innovation MMM can become a financial tool to guide
media Brand’s investments 
Myths
JUNE
AUGUST
SEPTEMBER

11. Case studies
10
thanks to new advancements
in technologies, modern solution
allow brands to analyze the MMM
incrementality measurement at a
daily basis.
“MMM is not real time”
Myths
MEASURAMENTS
Last month
Last few days
Last year
Last 3 years
2 days ago yesterday

12. Application
11
Incrementality tests serve the purpose of
validating the incremental contribution of
each media and to calibrate the MMM.
Incrementality testing methodologies are:
The MMM accuracy
depends solely on the
quality of historical data.
In most cases before it’s used
for media allocation decisions
it needs to be calibrated.
Geo-Lifts Conversion
Lifts
Audience
A/B Tests

13. Application
12
Ensure data is complete
and accurate
Remove errors
and
inconsistencies
Adjust
parameters for
better accuracy
Test Incremental impact
on each media
Evaluate model’s
precision and
reliability
Make informed
decisions based on
results

14. Application
13
3 years of historical data
Your data needs to have:
Output
Sales Revenue
New customers
Conversions
Context Factors
 Events (boolean values)
Promotions (boolean values)
Discount (% values) External
variables (boolean or nominal
values)
Organic
SMS
Email
impressions
Media (online
and offline)
Spend
Impressions
data you need
Daily or Weekly granularity
Data should be split by country or
grouped into important regions
Data should be in tabular format, one
column per variable, one row per date
13

15. Application
14
Diminishing
Returns
Revenue
Invested Budget
AdStock Effect over Time
Effect
Time
Decide to aggregate or
remove variables based
on these data quality
check 
Define Adstock &
Diminishing returns
hyperparameters
Select variables for the
models Detect if your data has:
Multicollinearity
Not enough
records 
not enough
Spend share
Calibration through priors
(Not required)
If you have prior knowledge of a
channel incremental contribution.
Add it before the training:

•Choose the channel
•Select the time
window
•Metric: Incremental
Sales Detected
(We suggest to use only reliable
incrementality sales data detected by
running an incrementality test) 

16. Application
15
Assess uncertainty with confidence intervals around your
estimated iROI
Ms Dv
Ms Fb Ecom
Tik Tok
Ms Google Nonbrand
Ms Groovin
Ms Rtb
0 20 40 60 80 100 120
iROI
Wider they are, the more uncertainty there is.
Run experiments on the most uncertain

17. Application
16
Refresh your MMM and calibrate it with our
incrementality test data monthly
2019 - 2023
TEST MODEL
Actual vs Predicted Revenue

18. Application
17
Simulate scenarios with your media
mix and predict future outcomes
current spending
optimal spending
Optimal Total Advertising Budget
€ 10,000,000.0
Optimal Total Incremental (paid)
€ 250,00,000 Optimal Total (paid) ROI
21.11

19. Thank you
Contact us at [email protected]