Overview
Cross-channel advertising numbers answer different questions at different scales: MMM anchors weekly channel contributions, IMA allocates them to daily campaigns, and randomized experiments supply causal reference points. Privacy loss can widen compatible effects, so certification depends on bounds, thresholds, and granularity. The series shows how shared semantics preserve meaning without turning modeled allocations into causal proof.
Learning outcomes
- MMM and IMA answer different resolution and planning questions; IMA adds campaign detail under a channel-level modeled anchor.
- A campaign allocation can reconcile to a weekly channel contribution without proving each campaign’s independent causal effect.
- Privacy loss can create a range of compatible clean effects, so certification requires bounds, thresholds, and evidence status.
- Finer reporting may increase actionability while worsening noise, suppression, multiplicity, and heterogeneity; granularity must match the claim.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
How do MMM and IMA differ in cross-channel advertising measurement?
MMM estimates weekly channel-level contributions for strategic planning, while IMA allocates those modeled contributions to daily campaigns within each channel. IMA preserves the aggregate anchor and carryover structure, but its campaign output remains a model-based allocation rather than independent causal proof.
Does reconciling campaign results to a channel total prove causality?
No. Reconciliation shows that a granular allocation preserves its modeled channel target, not that each campaign caused its allocated amount. The reported representative-channel fit supports reconstruction consistency, while independent campaign causality would require stronger evidence such as a suitable randomized experiment.
What does privacy loss change in incrementality measurement?
Privacy loss can remove matches, identity links, out-of-window conversions, sparse cells, or reporting precision, so a released report may support a range of clean effects rather than one identified lift. Retention, suppression, noise, and segment heterogeneity determine how wide that range is.
How should reporting granularity be chosen for an advertising decision?
Choose the partition that matches the decision, outcome, time window, and threshold while surviving noise, suppression, multiplicity, and heterogeneity checks. Coarser cells improve stability but can hide differences; finer cells add specificity but may leave claims unresolved.
When is a privacy-affected advertising result certified, rejected, or unresolved?
Certify when the finite-sample lower bound exceeds the business threshold, reject the above-threshold claim when the upper bound is at or below it, and mark it unresolved when the compatible interval crosses the threshold. The status depends on the stated outcome, threshold, and ambiguity set.