A booking date is not an exposure date
Podcast media can be purchased ahead of release and heard over a longer period. A weekly invoice series may therefore be poorly aligned with when listeners encounter the message. Before fitting an MMM, identify what the spend and delivery timestamps represent.
- Record actual delivery
- Map release and listening lag
- Separate code attribution
- Test a bounded decision
Keep booking, publication, delivery and billing fields separately where available. The data requirements guide helps establish a consistent contract before choosing a lag transformation.
A useful model starts with a consistent outcome and a decision it can inform. The marketing mix modelling learning path connects this idea to the assumptions, practical checks and next steps.
To work through the connected decisions, read Adstock half-life: convert decay into a time assumption and MMM channel grouping: when separate ROAS is unsupported.
Distinguish listening delay from effect persistence
An episode released on Monday may continue to receive downloads over several weeks. Advertising influence may also persist after listening. Those mechanisms can both create a delayed sales pattern, but a fitted adstock parameter does not automatically distinguish them.
If exposure estimates are available by week, align the model to those estimates and document their limitations. If only publication dates are available, state the delivery assumption. The adstock guide explains why time units and the carryover window matter.
Use promo codes as one signal
Suppose a fictional campaign records 200 code redemptions. Some customers may have bought without the ad, some exposed customers may buy without the code, and codes can be shared beyond the intended audience. The redemption count is therefore neither a complete count of influenced customers nor a direct estimate of incrementality.
Reconcile code sales to business revenue and record the offer's discount cost. Keep attribution and causal estimation separate. A code can still be useful for diagnosing response timing and delivery problems without being treated as a perfect measurement system.
Design a decision the evidence can support
A small number of national podcast bursts that coincide with other campaigns may be difficult to separate in an MMM. Inspect the calendar and compare plausible lag assumptions. If the effect is unstable, consider a staged schedule, a suitable holdout or another design that creates more informative variation.
For the next budget decision, specify the audience, shows, creative, delivery window and commercial outcome. A response estimated for one show mix may not transfer to a different audience. Use the testing guide to define the contrast and uncertainty.
The review pack should show delivery evidence, timing assumptions, code activity, business outcomes and the range of model recommendations. A strong narrative about listener trust is not a substitute for an observable, bounded test of the proposed spend change.
Sources and further practice
This article’s worked example is illustrative, not a measured client result. Use the linked methodology to check the assumptions before applying it.