[GA4] Trend change detection

Trend change detection surfaces subtle but long-lasting and important changes in the direction of your data. It's similar to anomaly detection in that they both detect a change in data. The main difference is that anomaly detection highlights sudden spikes or dips in data, while trend change detection highlights subtler changes over a longer period of time.

Google Analytics displays these changes in an Insight card in the Insights & recommendations section of the Home page, in Reports snapshot and Advertising snapshot, and in the Insights hub.

Understanding trend changes

Many types of changes can lead to a trend change. Some changes are predictable and intended. In these cases, you won't need to take any action. For example, if you start a new advertising campaign, you might see a gradual increase in revenue; Analytics detects the trend change as the start date of the campaign.

Other changes are unexpected and require more attention, such as important metrics that trend downwards. For example, due to a recent update to your website code, the registration button on your website stops working. You may not notice an immediate drop in active users, but over time, your growth rate stalls as fewer users are able to register. Analytics will detect this change so you can make any necessary adjustments to your site.

How it works

When Google identifies a change in the direction of time-series data for a given metric, Google overlays a circle on the date when the change occurred. You can hover over the circle to learn more about the change, including the previous and current rate of change and when the trend change occurred.

You can also click Investigate report at the bottom of the Insight card to further investigate the data. Clicking the button opens to a more detailed report with the same date range, metric, and dimension applied. You can adjust the date range, compare other dimensions, or add more dimension breakdowns to investigate the trend change.

Types of changes detected

Trend change detection surfaces a trend change when the forecasted value differs from the actual value, which happens when a trend changes from:

  • An increase to a decrease
  • A decrease to an increase
  • A greater increase or decrease
  • A lesser increase or decrease

Troubleshoot trend changes

Before you make any changes to your website or app when a trend change occurs, it's important to first consider whether it's an expected change and what the cause is if it is unexpected. If the change is expected or positive, you likely don't want to make any changes.

The following describe possible reasons why a trend change occurred:

Business cycles

Traditional business cycles can lead to a negative trend change that is expected and usually doesn't need extra attention unless the trend is out of a reasonable range. Examples include end of holiday seasons for ecommerce websites and start of school seasons for gaming apps.

Changes to the Analytics property

Configuration changes to your property could lead to a trend change. In Admin, under Property, click Property change history, and check if any changes occurred before the trend change. Learn how to View the history of account/property changes

Changes to your website or app

A negative trend change can happen if you didn't set up your website or app correctly. Examples include:

  • Technical issues like increased latency (or page loading times), server overload, etc.
  • Missing or wrong settings in your pages’ metadata, like robots.txt, sitemap.xml, SSL, etc. Search Engines use metadata to rank your page and misconfigured metadata can lead to trend change in organic traffic.
  • Ads campaigns are set ineffectively, when the trend change happens only in paid traffic.

How Analytics detects trend changes

Analytics Intelligence applies a signal segmentation algorithm to a time series to detect whether a trend change has occurred. The model segments the time series into several parts so that in each part, the data shares similar patterns while different parts have different patterns. The boundary points between two parts become the trend changes that we report.

The typical training period for detecting changes to daily data is approximately 90 days and weekly data is approximately 32 weeks.

Was this helpful?

How can we improve it?
Search
Clear search
Close search
Google apps
Main menu