Invalid traffic & fraud · also called Anomaly detection
Data anomaly detection
Data anomaly detection is spotting unusual patterns in traffic or campaign data, such as sudden spikes, impossible ratios or odd timing, that suggest fraud, errors or broken tracking.
Most fraud leaves statistical fingerprints. Anomaly detection learns what normal looks like for a site, app, supplier or campaign, then flags deviations: traffic spikes at 4 a.m., a sudden jump in click-through rate, far more impressions than page views, conversions arriving in perfect intervals, or a new device-ID share that is far above normal.
Methods range from simple thresholds and ratios to time-series models, clustering and machine learning that compares each source against its peers. In SIVT detection, anomaly detection provides the population-level view that single-event checks miss: one session may look fine, but ten thousand identical sessions do not. The MRC IVT guidelines expect measurers to analyse traffic for such patterns rather than rely only on fixed lists.
Not every anomaly is fraud. A breaking news story, a TV mention, a holiday sale or a tracking bug can all create spikes. Good practice is to investigate anomalies, corroborate them with other signals like device fingerprinting and IP reputation, and document the conclusion.
Think of it like this
It is like a bank noticing your card being used in three countries within an hour: not proof of theft alone, but definitely worth a call.
An example
A publisher's dashboard alerts on one ad placement whose CTR jumped from 0.3% to 6% overnight, all from one mobile carrier; it turns out a sub-publisher had started running accidental-click interstitials.
Related terms
Fraud detection
Fraud detection in advertising is the process and technology of identifying invalid or fraudulent impressions, clicks, installs and conversions, so they can be blocked in advance or excluded from billing.
Behavioral analysis
Behavioral analysis in fraud detection is examining how a visitor interacts, such as mouse movements, taps, scrolling, typing rhythm and session timing, to tell real humans from bots and scripted activity.
SIVT (sophisticated invalid traffic)
SIVT (sophisticated invalid traffic) is invalid traffic designed to look human or legitimate, which can only be found through advanced analytics, multi-point corroboration and often human review.
Discrepancy
A discrepancy is the difference between numbers reported by two systems for the same campaign, such as a publisher's and an advertiser's ad server counting different impressions or clicks.
Traffic quality
Traffic quality is how genuine, attentive and valuable a source of visitors or ad impressions is, combining invalid-traffic levels with signals like viewability, engagement, brand safety and conversion.
Sources: MRC Invalid Traffic Detection and Filtration Guidelines Addendum (2020 update)