pct_of_spend parameter

[arodriguezmio]

Hi everyone,

I have a question regarding the pct_of_spend parameter in Meridian.

As I understand it, when pct_of_spend is specified, it effectively fixes the allocation according to the historical spend distribution. In my case, since I configure it to match the historical split, the optimization report shows a variation of zero — which makes sense because I am constraining it that way.

However, my question is:

Is there a way to evaluate what would happen if I enforced a different investment split (i.e., a custom allocation) and compare its impact against the historical distribution?

What would be the recommended way to do this in Meridian?

Thanks in advance!

[gousesb]

Hello @arodriguezmio ,

Thank you for contacting us!

Yes, you can evaluate custom investment splits against historical distribution in Meridian using the pct_of_spend parameter in the BudgetOptimizer.optimize()method. The pct_of_spend parameter defines the non-optimized baseline allocation across all media and reach-and-frequency (RF) channels. It consists of a numeric list of size n_paid_channels where the order must match the sequence of InputData.media + InputData.reach. Each value in the list must be between 0 and 1 per channel, and the entire list must sum to 1.0 (100%). By default, if this parameter is unspecified, it defaults to the historical spend proportions. Budget and allocation are used in conjunction to determine the non-optimized media-level spend as budget * pct_of_spend, which is then used to calculate the "Non-optimized" performance metrics (e.g., ROI).

The spend_constraint_lower and spend_constraint_upper arguments play a critical role in defining the "feasible range" for the optimizer by setting the lower and upper bounds of media-level spend based on the baseline. The optimizer uses the baseline spend to determine these constraints for each channel; for example, the lower bound of media-level spend is calculated as (1 - spend_constraint_lower) * budget * pct_of_spend), while the upper bound is (1 + spend_constraint_upper) * budget * pct_of_spend). By setting these constraints, you tell Meridian how much it is allowed to shift spend away from your defined pct_of_spend. Your zero-variation occurred because identical lower=upper constraint values created zero-width feasible ranges for each channel, completely locking the spend allocation exactly at the baseline, leaving no room for the optimizer to shift funds between channels.

The following sample code demonstrates how to set up both a historical baseline and a custom allocation for comparison:

# Sample code for evaluating custom investment splits


build_channel_args = meridian.input_data.get_paid_channels_argument_builder()

# Historical baseline (Matches your historical spend distribution)
pct_of_spend_hist = build_channel_args(Channel0=0.4, Channel1=0.2, Channel2=0.2, Channel3=0.2)

# Custom allocation to test (A different strategic split)
pct_of_spend_custom = build_channel_args(Channel0=0.3, Channel1=0.3, Channel2=0.2, Channel3=0.2)


# Constraints define the feasible range.
# Setting these to 0.3 allows +/- 30% movement from the pct_of_spend baseline.
# Setting these to 0.0 would lock the optimizer to exactly the pct_of_spend provided.
# Defaults to 0.3 for fixed budget and 1 for flexible budget.

spend_constraint_lower = build_channel_args(Channel0=0.3, Channel1=0.3, Channel2=0.3, Channel3=0.3)
spend_constraint_upper = build_channel_args(Channel0=0.3, Channel1=0.3, Channel2=0.3, Channel3=0.3)

# Run 1: Historical baseline evaluation
opt_hist = budget_optimizer.optimize(
      budget=70_000_000,
      pct_of_spend=pct_of_spend_hist,
      spend_constraint_lower=spend_constraint_lower,
      spend_constraint_upper=spend_constraint_upper)

# Run 2: Custom allocation evaluation
opt_custom = budget_optimizer.optimize(
      budget=70_000_000,
      pct_of_spend=pct_of_spend_custom,
      spend_constraint_lower=spend_constraint_lower,
      spend_constraint_upper=spend_constraint_upper)

See Budget optimization scenarios for more details.

By comparing the results of opt_hist and opt_custom summaries, you can evaluate which investment split is predicted to yield a higher incremental outcome or ROI based on the model's learned response curves.

Feel free to reach out for any further queries or feedback regarding Meridian.

Google Meridian Support Team