time period setting for budget optimization

[oneorange101]

Hi everyone!
I have built a meridian model and its training data period was [2023.6 ~ 2025.6] (this is our maximum data for now...).
Using the model, I want to get the optimal media mix for our next campaign that is scheduled in [2026.2 ~ 2026.6].

As you know, median allow us to select specific time period to run an optimization within the training data period, so i am wondering how I can select the proper time period.
In this situation, what would be the best choice?

1. running optimization for full data period [2023.6 ~ 2025.6]
2. running optimization for [2025.2 ~ 2025.6]
   (the most similar target period in the training data)

It seems that 1) could be the most comprehensive allocation result using whole data, but 2) seems also reasonable.
Could you please provide any guideline?

Thanks in advance for any help.

[cpulavarthi]

Hello @oneorange101,

Thank you for reaching out!

Between the two options you presented, Option 2 (running optimization for [2025.2 ~ 2025.6]) is generally the better approach if you feel that this period most closely resembles the period over which you are setting budgets. In this case, keep in mind that while the optimization procedure considers only a subset of the data, the model training process still takes advantage of the entire dataset.

Alternatively, if you have the resources and data, you can use the new_data argument and pass the expected forecasts of media_spend and the corresponding media exposure metrics to get most relevant budget allocation guidance for the future time period. To do this, you would create a dataset for the [2026.2 ~ 2026.6] period and then apply budget optimization over this new period.

Feel free to refer to our documentation on Budget Optimization which may help you better understand this process. Do check out the page on scenario planning and future budget optimization which is most relevant to your current question.

Do reach out if you have any further questions or suggestions regarding Meridian.

Google Meridian Support Team