How does Meridian integrate media_spend and media activities together?

[mikeh3636]

Hi Meridian support team,

Both media spend and media activity metrics (e.g., impressions, clicks, etc.) are required as inputs. Could you clarify how the model integrates these two types of metrics? Do they both undergo the same transformations, such as Adstock, Hill functions, and population scaling? Additionally, how does cost interact with impressions to standardize the modeling process?

Thank you.

[AdimDrewnik]

You can use spend variables twice.

[mikeh3636]

yeah, but it is not the recommended approach.

[AdimDrewnik]

Not recommended by Meridian but recommended by Robyn, so in the MMM world same thing can be at the same time recommended and not recommended.

[cpulavarthi]

Hello @mikeh3636,

Thank you for contacting us!

Meridian only uses spend data to calculate the ROI denominator. This comes into play when setting ROI priors, and also for post-modeling analysis such as budget optimization.

Otherwise, all modeling is based on the media units. Only the media units undergo population scaling, Adstock, and Hill transformations.

Media units can be spend, impressions, clicks, or any unit you choose. This can vary by media channel, and there is no specific recommendation. Because media is a treatment variable, the unit should be something that is intervenable (in other words, the advertiser should have a reasonable degree of control over it).

Feel free to reach out for any further queries.

Thank you

Google Meridian Support Team

[kacper-sellforte]

Response curves are calculated with spend on X axis though. That means that there needs to be assumed relationship between spend and modelled metric, right? Is there a constant scaler being used to go from spend to modelled metric and the perform transformations on this modelled metric to obtain response curve?

[AdimDrewnik]

I think I saw in the documentation that it is a constant scalar approach, i.e. assumption of e.g. constant CPC throughout 2-3 years.

[cpulavarthi]

Hello @kacper-sellforte,

For each point on the response curve, Meridian takes the ratio of the x-axis value divided by the total historical cost. This ratio is used to scale the media units in each geo and time period (keeping the historical flighting pattern in tact and just scaling everything up or down). For example, if the total channel spend were $100, then to plot the point at x=$50, the media units in each geo and time period would be scaled by a factor of 0.5.

You could think about it as assuming a different constant cost per media unit within each geo and time period, where the constant is the historical average cost per unit in each geo and time period. Equivalently, you could think about it as assuming the same cost per media unit for all geos and time periods, where the constant is the historical average cost per unit overall. Either way, scaling all media units by 0.5 results in a total cost of $50.

The details are explained in our documentation linked here.

Feel free to reach out for further queries.

Thank you

Google Meridian Support Team