There are numerous approaches to estimating these effects, such as analyzing panel data, regression analysis or running experiments. This provides a far more nuanced measurement of the impact that accounts for the differences across channels and ad types, and in purchase or consideration cycles across categories of products as well.
This is particularly important for categories with longer consideration periods like electronics, or in channels which are generally used for brand-building rather than direct response advertising. Adstock can better capture the impact of retail media as it moves into new media beyond search.
Measuring adstock
The basic formula is: Adstock = Previous Adstock + (Advertising Exposure × Decay Rate).
- Previous Adstock is the adstock from the previous period—most often measured in days, but sometimes weeks or months. If you are starting with the first period, this number would be zero.
- Advertising Exposure is the level of advertising exposure in the current period. This could be measured as impressions or another relevant advertising delivery metric.
- Decay Rate is the rate at which the impact of advertising diminishes over time. This is typically expressed as a decimal between 0 and 1. A number closer to 0 indicates a higher decay rate; a number closer to 1 indicates a slower decay with a greater buildup effect over time.
It’s important to note that determining the decay rate is crucial and may depend on various factors.
- Brand or category difference: Is this a category with a long consideration or research cycle?
- Product differences: Because retail media is often tied to supporting a specific SKU, differences in consumer behavior at the SKU level need to be accounted for. Is this a product people buy once a week? Once a month? Once in their life?
- Ad memorability: How engaging is the creative or ad unit? A more engaging format such as video may generate a greater likelihood of someone recalling it later.
A combination of regression analysis on historical campaign and sales data, plus experiments like randomized controlled tests or geo-matched market tests, is typically used to estimate this decay rate. In more advanced applications, there are considerations around potential data transformations, like applying a Weibull distribution instead of a geometric one to better account for the shape of the curve and scale of spend for some media.