Anomaly engine

In shortThe Anomaly engine checks whether your most recent day of GA4 sessions fell more than 15% below the same weekday in the previous two weeks. If it did, it splits the drop by source and medium, country and device, names the biggest contributor and writes a plain-English explanation you can send a client. Open Analytics → Google Analytics 4 → Anomaly engine and click Detect Anomaly. It flags drops, not spikes, and costs credits per run.

What the Anomaly engine is for

Traffic looks low today and someone asks why. The slow way is to open GA4 and compare sources, countries and devices one by one against last week. The Anomaly engine does that comparison in one click and tells you which slice explains most of the gap: one referral source that stopped sending visits, one country, or mobile traffic only.

Use it when a day looks wrong and you want a first answer fast. For slow declines over weeks or months, use Content decay radar and Algorithm impact on your Search Console data instead.

Before you start

• GA4 connected and the right property selected. See Connect GA4 and read the overview.
• At least two weeks of session data, so there are earlier days with the same weekday to compare against.

How it decides there's an anomaly

1. It reads daily sessions for the last 15 days (14 days ago up to today).
2. It takes the latest day in that data and finds the earlier days in the window that fall on the same weekday. In 15 days that's normally two: 7 and 14 days before.
3. It averages those days to get a baseline and works out the change.
4. If the latest day is more than 15% below the baseline, it calls it an anomaly and splits the drop by source and medium, country and device. Otherwise it reports No Anomaly Detected.

Run it step by step

1. Open the Anomaly engineGo to Analytics → Google Analytics 4 → Anomaly engine. The page is titled Root-Cause Anomaly Engine, with the price on a badge next to the button.
2. Click Detect AnomalyThe button shows Scanning... while it fetches the data and, if needed, writes the explanation.
3. If nothing is wrongA green card says No Anomaly Detected, with the latest day's sessions, the baseline and the percentage change. If there wasn't enough history, the card says so instead.
4. If a drop is foundA red banner reads Traffic Anomaly Detected on the date, with Sessions, Baseline and Drop. Below it, Primary Root Cause names the slice that explains the largest share of the drop.
5. Read the explanation and breakdownThe Client-Ready Explanation describes what happened in plain language, says whether it looks like a tracking problem or a real traffic change, and suggests one next step. Variance Breakdown has three tables, By Source / Medium, By Country and By Device Category, each with Anomaly Day, Baseline and Contribution.

How to read the breakdown

Contribution is the share of the total drop that one value explains. Values can add up to more than 100% across a table when some segments rose while others fell.

What it costs