Performance can be created in reporting
Fou revisits earlier click-fraud and attribution disputes to illustrate a broader point: a vendor may appear highly performant when its reporting claims sales that were likely to occur anyway.
Bots are not required
Non-incremental attribution can happen entirely with real people. A marketer who looks only at bot or IVT rates can therefore miss a major economic quality problem.
Several platforms can claim the same sale
Google, Meta, TikTok, DSPs and other vendors may each report conversions under their own attribution windows. Those numbers cannot simply be added because a single purchase can receive credit from several systems.
Incrementality asks the causal question
The relevant question is which sales or leads would not have happened without the advertising. Depending on the setup, answering that may require experiments, holdouts, control groups, baselines or other causal designs.
Separate traffic quality from effectiveness
Forensics asks whether delivery and traffic were real and plausible. Effectiveness analysis asks what additional business outcome that real delivery caused. Both are required, but they are different methodologies.
This article describes principles and questions. Specific fraud, bot or quality rates cannot be transferred directly to individual campaigns without understanding methodology, scope and data basis.