Experiment paths
Split a journey into 2 or 3 random paths, compare their conversions and revenue, and send everyone down the path that wins.
A Split step set to By percentage sends people down 2 or 3 random paths. Each path is a variant of the journey from that point: a push against an in-app message, a one-day wait against a three-day wait, a discount against none. The report compares the paths so you can keep the winner.
Set up a path experiment
In the journey editor, add a Split step and choose By percentage. Give each path a share. Shares add up to 100%, and each path needs at least 1% before launch. Add a path adds a third.
Add the steps that make each variant different. Paths can rejoin further down, so only the tested part needs to differ.
Path results measure the journey's first goal. Without one, paths can't be compared on conversions. See Set goals.
Each person reaching the split takes a random path by share, and always takes the same path if they reach that split again.
Read the results
Each percentage split gets a card on the journey's Results tab once someone has taken one of its paths. Paths are labelled Path A, B and C, in their order on the split.
| Column | Shows |
|---|---|
| Group | The path, with its share |
| People | People who took the path in the period |
| Converted | People who reached a goal within the first goal's window after taking the path, and the rate |
| Revenue | Their store purchases in the same window |
| Per person | Revenue divided by people |
| Confidence | How sure the conversion-rate difference from Path A is real |
| vs A | The difference from Path A |
A sentence above the table names the leading path: by revenue per person when anyone bought, otherwise by conversion rate.
- Only people who reached the split count. Anyone who left earlier, or reached a goal before the split, is in no group.
- Path A is the baseline. Other paths are compared with it. Path cards have no control group; the journey's control group is compared on the journey's goal card.
- Confidence needs people. It shows once both compared paths have at least 100 people, and covers conversions only, not revenue.
- The window is the first goal's window. Without a goal, revenue counts for 72 hours after the split.
- Changing shares later doesn't rewrite history: each person counts under the path they took.
Journeys that ran before experiment paths existed (2 October 2026) recorded no paths and have no path cards.
Send everyone down the winner
When a path leads, its card offers Send everyone here. Confirming sets that path to 100% and the others to 0%:
- Everyone reaching the split from then on takes the winning path.
- People already past the split stay where they are.
- You can change the percentages back in the editor.
If the lead isn't clear yet, the confirmation warns that there aren't enough people. A lead is clear when its conversion confidence is at least 95% and, if people bought, the top earner per person also converts best. With three paths, confidence is only measured against Path A, so a lead by B or C is never treated as clear: a close third path can't be ruled out.
You can also set percentages yourself in the editor. After launch, a path can go to 0% as long as the others add up to 100%. Paths can't be added, removed or reconnected after launch.