Define the population before comparing the result
AMC measurement connects an exposure or customer group to an outcome. The result depends on who entered the group, what counts as an outcome, and how much time conversions had to arrive. Align those definitions before comparing campaigns or drawing a Sankey chart. Otherwise, a difference in the reporting setup can look like a difference in performance.
| Measurement question | Definition that must stay explicit | Deeper guide |
|---|---|---|
| How many new customers did we reach? | Distinct new-to-brand customers, separate from orders, units, and revenue | New-to-brand customer metrics |
| What exposures preceded conversion? | Ordered exposures, campaign grouping, conversion lookforward window | Path-to-conversion analysis |
| How do DSP and sponsored ads overlap? | Mutually exclusive exposure groups and matching campaign scope | Four-way overlap analysis |
| How does subscriber spending differ? | Subscriber events, repeat orders, and comparison population | Subscribe & Save analytics |
| What happened outside Amazon? | Configured Events Manager signals associated with the relevant DSP order | Off-Amazon conversion measurement |
Does an overlap or spend gap prove incremental impact?
No. An exposure-group comparison or subscriber spend gap describes the populations included in the analysis. It does not, by itself, show what those same customers would have done without the campaign or subscription. Keep the observation separate from the causal explanation and the proposed action.
The overlap guide addresses double-counted exposure; the Subscribe & Save guide distinguishes a spend gap from proof of cause. Both can inform investigation without becoming an unsupported claim of incremental revenue.
Why can the same analysis change after it first runs?
Conversions can arrive after the exposure window, and privacy aggregation can suppress sparse groups. In path analysis, grouping campaigns helps preserve interpretable paths; waiting for the intended conversion window makes successive reads comparable. A day-zero chart and a later chart may have different completeness even when the query is unchanged.
For off-Amazon outcomes, check that the signals are configured and available before treating missing events as no conversions. Keep purchase sales separate from non-purchase conversion value. Each reporting artifact should retain its population, source, date window, and known gaps beside the result.
From a measurement result to the next decision
These guides cover selected measurement questions, not every AMC attribution model. Use AMC SQL and advanced modeling for source selection and validation, and AMC audiences when a reviewed finding becomes an audience proposal. The AMC introduction covers access and fundamentals; Kuudo's AMC software connects the analysis to a repeatable operating workflow.