Skip to content
GetProfit
Log in

← All terms

Anomaly detection

Anomaly detection is the automatic search for figures in an ad account that break sharply from their usual level, such as a sudden drop in sales.

How it works

A detector compares a current figure with a baseline from the account’s own history and sends an alert when the gap passes a threshold you set. Google’s Account Anomaly Detector script, for example, compares today’s impressions, clicks, conversions and spend so far with the average for the same weekday in past weeks. Google’s example uses 26 weeks. It skips the last 3 hours: statistics can arrive that late.

Results swing on their own, so a threshold tighter than the account’s usual swing raises false alarms. Fast checks catch breakages, such as a conversion tag that stops firing or a bestseller that drops out of impressions. Slow checks catch a decline no single day shows.

Our guides cover anomaly alerts without false alarms and bestseller alerts.

Formula

(Current value − baseline) ÷ baseline, compared with the threshold

Where you see it

  • Google Ads: scripts such as the Account Anomaly Detector, and Explanations, which say why performance changed significantly.
  • Portal: the products screen lists products that used to bring money and dropped out of impressions.

Example

Example store, not client data.

By 4 pm on an average Tuesday, the tableware shop has 170 clicks. This Tuesday it has 85: (85 − 170) ÷ 170 = −50%. The threshold is −30%, so the script sends an alert.

How GetProfit reads it

The portal treats a product’s conversion value for one day as anomalous when it is more than 30 times the account’s typical order value. With a typical order of 600, that is anything above 18,000. Such conversions stay out of the portal’s figures. The conversion check lists them so you can confirm or retract each one.

Not to be confused with

  • Explanations — a Google Ads feature that suggests why a change happened. Anomaly detection finds the change.
  • Change history — the log of edits in the account, which shows possible causes rather than deviations in results.

Benchmarks

In GetProfit data (110 stores with at least 8 months of history, June 2025 – June 2026), the median store’s monthly ROAS typically differs from its own median by 16% (middle half of stores: 11.0–24.1%). Your niche may differ.

Sources