Adstock Decay Rate Query

[annie-gulati]

We tried checking for the adstock decay rate for our model with max lag of 6. Here are the results for few of our channels (see table below):

We have following queries around it

1. Can you please help us explain how it is actionable to client
2. If some variable has adstock value as for example 0.55 in channel A, what is the right way to communicate that.
3. Is there a way to assign individual lag values for media channels

<style> </style>

Channel	Time unit	Mean
Channel A	0	1
Channel A	0.2	0.958483268
Channel A	5.8	0.552175676
Channel B	0	1
Channel B	0.2	0.907831595
Channel B	5.8	0.311651142
Channel C	0	1
Channel C	0.2	0.85775932
Channel C	5.8	0.186303556
Channel D	0	1
Channel D	0.2	0.772829091
Channel D	5.8	0.051461596

[cpulavarthi]

Hello @annie-gulati,

Thank you for contacting us!

Adstock decay value typically represents the percentage of ad’s impact that remains over time. It accounts for lagged effect (carryover), which is the delayed impact of advertising on sales. To answer your questions:

1. Adstock decay rate can be used to explain how the target KPI is being impacted by current ad-spend in a subsequent (future) time period. This can help devise marketing strategies along with information from the Hill function which gives the amount of incremental increase in returns for each additional unit of ad-spend. This information can be used to devise how to allocate marketing investments.

2. If a particular media channel has an adstock value of 0.55 at a point in time, this means that the impact of the spend on that channel diminishes to 55% of the initial impact in the subsequent time period. As an example for easier understanding, if the initial impact at a particular point in time is 100 units and the adstock decay is 0.55, then in the next period the estimated impact would be 55 units.

3. Meridian doesn’t allow assigning individual lag values to each media channel. The max_lag parameter captures the maximum value of lag present in all channels. If a particular channel has a lower carryover effect and thus lower lag value, the model will simply train and assign the coefficients accordingly so that this is captured. Setting max_lag value as the maximum possible lag for all channels is the recommended approach. You may check our documentation on Setting the max_lag Parameter for more information regarding this.

Additionally, you may use the plot_adstock_decay() method if you wish to visualize the evolution of adstock decay rate over time (Ref - Adstock Decay).

Feel free to reach out for any further queries regarding this.

Thank you

Google Meridian Support Team