Email Marketing Channel Integration in MMM Analysis

[SteliosSake]

Hello there!

I am currently in a setting with a national model, utilizing 2.4 years of data of weekly granularity (125 weeks). The configuration is the following:

• KPI: Profit (Net, not Gross)
• Paid Media: 12 Channels (the spend of each channel is utilized for both the "Media Spend" and "Media Data" variables. because we don't have impressions or clicks available for each channel)
• Non-media: Discounts
• Knots: 56 knots placed densely during holiday seasons

I want to add at this point the Emails marketing channel because It is important but I need some guidance regarding its integration as the 13th marketing channel.

Things to consider:

• The spend of the email channel is fixed every week (a service is utilized that manages all our emailing), which leads to a lack of variability in the treatment variable that Meridian utilizes for calculating the incrementally of the channel (if I am not mistaken). If I choose the spend of the email channel for both "Media Spend" and "Media Data" variables, will this create a problem for the model due to the lack of the variability in email spending?
• In the scenario that we don't want to use email spend for both "Media Spend" and "Media Data" variables, would it be better to utilize email sends, email opens or email clicks (somebody clicking a link inside the email and get redirected in the client's site) for the "Media Data" variable? Which one is the best choice in your opinion?

Thanks in advance for all your help!

Kind regards,
SteliosSake

[dattatreyam23]

Hi @SteliosSake,

Thank you for contacting us.

To integrate email marketing as a channel in your national model, you may consider a few steps, depending on your exact use case.

• You may use email impressions, clicks, or both as exposure variables if they best capture marketing pressure or engagement (not just spend)—often as a preferable media input. [See here]

• Prepare clean, weekly, geo-level data for each metric.

• When configuring your model, specify which metric(s) you use per channel; Meridian supports multiple exposure variables per channel when justified, but avoid high multicollinearity between clicks and impressions—pick the one that best represents incremental exposure for your use case. [See here].

• Include appropriate control variables (seasonality, promotions, overlapping campaigns) to ensure causal attribution is robust and to properly isolate email's effect.

• After fitting your model, validate that ROI and the estimated impact for clicks/impressions are plausible and in line with expectations; adjust variable selection or priors if needed.

• Additionally, I would also ask you to address the "lack of variability" concern with a way to mitigate it. Constant media spend shouldn't be an issue when fitting Meridian.

You may choose your best option, depending on your exact use case. Feel free to reach out if you have any questions or suggestions.

Thank you,

Google Meridian Support Team

[SteliosSake]

Hi @dattatreyam23 ,

Thank you for your quick reply! Your feedback is very important for my work!

I have the below follow-up questions on the points you raised:

"Additionally, I would also ask you to address the "lack of variability" concern with a way to mitigate it. Constant media spend shouldn't be an issue when fitting Meridian."

• So all in all you are saying that there is no reason to worry regarding the model fit if there is no variability in spending, correct? The fixed value that is the same for all weeks should be okay to be the input for the "Media Spend" variable.

"After fitting your model, validate that ROI and the estimated impact for clicks/impressions are plausible and in line with expectations; adjust variable selection or priors if needed."

• Regarding this point, in your opinion it is better to either:
  a.) have the same ROI priors for all media channels (treating them all as being at the same level of effectiveness in terms of ROI) and see if after the model fit the results of the model (the ROI estimations) align with the company experiments about ROIs?
  b.) take the company experiments (estimated ROI value of each channel) and define the priors based on that?
  Which method is better, a.) b.) or both?

"You may use email impressions, clicks, or both as exposure variables if they best capture marketing pressure or engagement (not just spend)—often as a preferable media input."

• I have impressions only for certain marketing channels, is it okay to have impressions only for a small subject of all the channels? Meaning that the "Media Data" variable will be impressions for 2 channels and spend for the rest of the 10 channels.

Thanks in advance for all your help!

Kind regards,
SteliosSake

[dattatreyam23]

Hi @SteliosSake,

Thanks for the follow-up. To answer your questions:

• Lack of variability: If a media spend variable is constant across all weeks and is only used in ROI calculations (not as a predictor in the model), this will not impact model fit in Meridian. No additional mitigation is needed in that case.

• ROI priors: We recommend using company experiments (estimated ROI per channel) to define priors. This approach ensures the model’s starting assumptions reflect real-world performance, and you can still validate post-fit results against those same experiments to check alignment.

• Mixed exposure variables: It’s fine to have impressions for only some channels and use spend for others. Meridian supports different exposure variables per channel—select whichever metric best represents incremental effect for that channel.

Hope this helps clarify. Feel free to reach out if you have any questions or suggestions regarding Meridian.

Thank you,

Google Meridian Support Team

[SteliosSake]

Hello @dattatreyam23 !

Thanks for all the feedback, I am 90% covered!

Could you elaborate more on the following statement:
"Lack of variability: If a media spend variable is constant across all weeks and is only used in ROI calculations (not as a predictor in the model), this will not impact model fit in Meridian. No additional mitigation is needed in that case."

Especially in the part "and is only used in ROI calculations (not as a predictor in the model)".

How is this connected to the parameters defined in the model (e.g., how do I ensure that the media spend is utilized only in the ROI calculations)? Could you provide a simple example for reference in your answer?

Kind regards,
SteliosSake

[cpulavarthi]

Hello @SteliosSake,

Thank you for the feedback! I'm happy to elaborate on our earlier statement regarding why lack of variability in media spend shouldn't be an issue.

The core of the concept lies in how Meridian is designed to differentiate between a media channel's effect (media) and its cost (media_spend). The reason why lack of variability in media spend won't be an issue is because media exposure metric is used in modeling the causal relationships, not the media spend. Media spend on the other hand is only used for ROI computations, and consequently the budget optimization.

Please check a previous discussion #530 which elaborates the difference between the media and media_spend arguments and how they are used in modeling. This is essentially how Meridian uses these variables, and you don't need to specify anything to ensure that the media spend is only used in ROI calculations.

Hope this helps. Do reach out if you have any further questions!

Google Meridian Support Team