Allowing for billboard sped (and its timing) in Robyn

[Dylan-Evans91]

Hi all

I am currently building a model that predicts new business sales (revenue) and I would like to feed in billboard spend, as this is one of the main strategies employed within our industry. The one thing that I am grappling with is how to make allowance for the timing of the spend. With digital spend, we have constraints in place to ensure the same ad is only shown to users a handful of times. This makes me assume that the relationship between spend and impressions is quite stable - which should allow the model to detect the underlying signal in the data.

In contrast, billboards present a different challenge. They generally remain up for several months, with payments made monthly. The first month of spend for a particular billboard generates many impressions, but in subsequent months, the same audience (typically the same people driving past) sees the ad repeatedly. Currently, I am including a record for billboard spend on the 1st of each month when payments are made. However, I worry that this approach might hinder the model's ability to accurately capture the adstock effects of billboards, and consequently, the overall adstock dynamics. This feels separate from a simple saturation issue, but I am happy to be challenged on that point.

I would greatly appreciate any input on how to better model this, especially if you have experience dealing with similar challenges. Specifically, I am interested in:

• Effective ways to adjust billboard spend data to reflect the diminishing gain we get from a user seeing the same ad multiple times.
• Best practices for supporting Robyn's ability to model adstock effects for billboards in comparison to digital ads.
• Any theoretical insights or practical examples from your own work.

Thanks everyone!