Test and personalization insights
The Insights tab in the composition editor shows how the variations of a single A/B test or personalization are performing, next to the content itself.
Where the tab appears#
Select an A/B test or Personalization component in a composition, or any variation inside one, and the property panel gains an Insights tab alongside Content and Context.
The tab is scoped to test and personalization components deliberately. Selecting an ordinary component shows no Insights tab, because there is no set of variations to compare. For project-wide numbers, including per-composition traffic, use the Insights dashboard.
If the tab appears with a "Not configured" message, the project has no tracking token yet. See activate Insights.
Choosing what to measure#
The Measurement criteria menu selects the metric the chart and the variation list are built from.
| Metric | What it counts |
|---|---|
| Visits | Sessions in which the variation was served |
| Views | Times the variation was rendered, including repeat views within one session |
| Bounce rate | Share of sessions that ended without further interaction after seeing the variation |
| All goals | Conversion rate across every signal configured as a conversion goal |
| A named goal | Conversion rate for that single goal |
The goal entries only appear once the project has conversion goals configured. Conversion rate is measured per session, so a visitor who converts twice in one session counts once.
tip
Start with a goal rather than visits. Visits tell you the traffic split is working; a goal tells you whether the variation is doing anything useful.
Reading the chart#
A date range control scopes the data, and a chart type menu switches between two views of it.
- Cumulative accumulates the metric over the period. Use it to read the current standing, since the lines settle as data accrues.
- Time series plots each day separately. Use it to spot when something changed: a deployment, a campaign, or a traffic source drying up.
Below the chart, each variation is listed with its value for the selected metric. The variation currently winning is marked with a trophy. For bounce rate and goal metrics, each row has a details popover breaking the rate down into the underlying session counts.
Clicking a variation in the list selects that variation's component in the canvas, so you can move between a number and the content producing it.
Comparison mode#
The Comparison mode switch replaces each variation's absolute value with its percentage difference against the control, and labels the control row as such.
For an A/B test the control is the variation designated as control. For a personalization it is the baseline. The switch is only offered when the loaded data can express a comparison, which means the metric has to be measured per session and there has to be a control to measure against. It is hidden for metrics like views.
Changing the control#
For an A/B test with more than one variation, the menu on a variation row offers Set as control, which recalculates every comparison against that variation instead. The control row then offers Reset to default control to undo it.
This only changes how the numbers are presented to you. It does not change the test, the traffic split, or what visitors see.
Test status#
A/B tests show a status callout above the chart. Personalizations do not, because a personalization has no winner to declare. Its variations are targeted at different audiences rather than competing for the same one.
| Status | Meaning |
|---|---|
| Collecting data | No sessions recorded yet for this test in the selected period |
| Test is running | Data is accumulating but no variation has separated from the control. The popover explains what specifically is missing: more sessions, more conversions, or a larger difference |
| Winning variant detected | One variation has outperformed the control with statistical significance |
| No control variant found | The test has no variation designated as control, so nothing can be compared |
Each status carries the sample size, p-value, and confidence level so you can see the evidence behind it. When a winner is declared, visitors keep being served according to the configured traffic split, and the result is not expected to change.
note
The winner shown here and the significance checkmark on the dashboard's active tests table apply slightly different rules. See how significance works.