Does the optimizer use roll-in adstocks when a time period is selected?

[Jon77Ruler]

Apologies, but I couldn't find the answer to this after some searching.

So the question is; if I select a time range that is equivalent to "last 8 weeks" for the optimizer; does it use adstock for channels from the preceding few weeks (e.g. 4 weeks of adstock from last 8 - 12 weeks) to increase response? Ergo - if I picked a different 8 weeks in time - with hypothetically exactly the same budgets - I would get a different optimization?

Or conversely; does the time period selected simply sets the respective channel budgets and all optimizations are considered based on no recent prior activity?

I'm comparing/contrasting the approach to that used by Robyn, where they calculate an average adstocked position and multiply the output by the number of time units chosen. i.e. the time period chosen very much changes the output.

[cpulavarthi]

Hello @Jon77Ruler,

Thank you for contacting us!

The carryover effect from previous time periods is considered while computing the incremental outcome in Meridian. Our documentation on the definition of Incremental Outcome in Meridian clearly shows how the lagged media impact is considered while computing the incremental outcome. This incremental outcome along with the expected outcome (defined as the sum of the baseline and incremental outcomes of all treatment variables) is used estimating the ideal budget allocation in Meridian (Ref - Optimization for media channels without reach and frequency data, Optimization for media channels with reach and frequency data). Therefore, Meridian does consider the carryover effect from previous time periods while optimizing the budget allocation.

To answer your first question, the optimizer may give different optimized allocations even with the same budget for different time periods depending on the incremental outcomes of the channels during the chosen period. It is not just because of the adstock considerations, but also because of the seasonality and trend, and all the other components which impact the incremental outcome. And the answer to the second question is no, the optimizer doesn’t assume no recent prior activity, it automatically accounts for the lagged effects.

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

Google Meridian Support Team