Overview

A single metric often hides meaningful differences between segments of your users. A change that lifts revenue overall might help users on one device type while hurting users on another, and an aggregate result averages those effects together.

Metric property breakouts let you split a metric by the values of an event property-such as country, device type, or browser. This shows you the lift and confidence interval for each value alongside the overall result. You configure the properties and values you care about once on the metric. Datadog computes a separate result for each value, using the same analysis as the parent metric.

A breakout is purely additive: the overall metric is still calculated across all events, and each property value is reported as its own row underneath it.

Configure breakouts on a metric

You define breakouts when you create or edit a metric. Breakouts are supported for metrics built on Product Analytics, Real User Monitoring (RUM), and warehouse data.

1. Navigate to the Metrics page in Datadog Product Analytics and create a metric or open an existing one.
2. Configure the metric definition.
3. In the Split by Metric Properties section, click Add Property.
4. Select a Property to break out by (for example, Country, Device Type, or Browser). The available properties come from the metric’s data source. For warehouse metrics, the property must be a column on the metric’s SQL model.
5. In the Values column, select the property values you want to break out (for example, CA, US, and UK).
6. (Optional) Click Add Property again to break out by additional properties, or click the trash icon to remove a property.
7. Click Save.

How property values are discovered

Datadog discovers the available values for each property automatically. It scans the metric’s data source over the previous 30 days and keeps the 50 most frequently observed values, ordered by how often they occur. This list refreshes once per day.

If you don’t see a value you expect, it might occur too infrequently to rank among the top 50 values, or it might not have appeared in the data source within the last 30 days.

View broken-out results

After an experiment runs against a metric that has breakouts configured, you can reveal the per-value results in the experiment’s results scorecard.

1. Open your experiment’s results and switch to the confidence-interval view.
2. Hover over a metric that has breakouts configured to reveal its row actions, then click the Split by Metric Properties icon.
3. Select the properties you want to break out by, then click Apply.

A Property column appears. The parent metric row is labeled Overall, and each selected property value appears as a sub-row (for example, Country: UK) with its own per-subject values, relative lift, and confidence interval.

To remove a breakout, open the metric’s actions again and click the hide action for that property (for example, Hide splits by Country).

Break out exploration charts

When you explore results in a chart, use the Split by selector and choose Metric property followed by the property name to plot each property value as a separate series.

How breakouts are calculated

For each property value you select, Datadog creates a sub-metric: a copy of the parent metric with an additional filter that keeps only the events where the property equals that value. For example, a breakout on Country = US analyzes the parent metric using only events where the country is US.

Each sub-metric flows through the same analysis pipeline as the parent metric, so every breakout value receives its own lift estimate, confidence interval, CUPED adjustment, and significance test, computed on its filtered subpopulation. The analysis method and confidence level configured for the experiment apply to every breakout.

Because each breakout analyzes a subset of the data, breakouts have smaller sample sizes than the parent metric. A breakout value may need more data than the overall metric to reach statistical significance.

Supported metrics and constraints

Metric property breakouts have the following requirements and limitations:

Note: Outlier handling (winsorization) configured on the parent metric is not applied to breakouts, because a threshold computed across the full population is not meaningful for a filtered subpopulation.

Further reading

Additional helpful documentation, links, and articles:

Create Experiment MetricsDOCUMENTATION Reading Experiment ResultsDOCUMENTATION Analysis MethodsDOCUMENTATION