A/B test two paths inside a flow
Divide traffic two ways — permanently, or until one path proves better.
Split the traffic sends people down two paths at random. It has two modes, and choosing the wrong one is the only real trap here.
The two modes
- Divide it — the split is permanent. Use it when both paths are meant to exist forever: half get the survey, or you’re holding back a control group.
- Test it — the split runs until one path is demonstrably better, then everyone goes that way. Use it when you’re trying to find out which email works.
It defaults to Divide it, so an existing split never starts picking winners on its own.
How a test decides
It waits until each path has had enough people — not until a deadline passes. That’s the important difference from a campaign A/B test, and it’s deliberate.
A campaign’s test slice is spent the moment it sends, so it has to hold back a remainder and wait a fixed number of hours. A flow never stops enrolling people, so there’s no remainder and no deadline — waiting costs nothing except time, and stopping early costs you a wrong answer.
The threshold defaults to 100 per path and can’t go below 20. Below about a hundred, a difference between two paths is very hard to tell from chance.
What it measures
Pick the metric that decides the winner:
- Clicks — the default, and the right answer most of the time.
- Goal — the strongest, when your flow has a goal. It measures the outcome rather than a proxy for it.
- Opens — available, but weak. Apple Mail opens messages on the recipient’s behalf, so an opens-based winner is close to a coin flip.
Only what happens after the split counts
A click on an email sent before someone reached the split is ignored. Otherwise a flow where most engagement happens early would look decided when it had learned nothing.
After a winner is picked
Everyone new goes down the winning path immediately, and the winning branch is marked on the canvas. People already on the losing path carry on through it — they’re mid-flow, and yanking them sideways would be stranger than letting them finish.
Test again clears the decision and reopens the test. It keeps the traffic counted so far, because the usual reason to re-test is that you changed one of the paths — not that the earlier traffic never happened. One consequence worth knowing: on an unchanged flow, re-testing will often re-decide almost immediately.
Common issues
The test never decides
Not enough people have reached it yet, or neither path has produced a conversion. The card shows how many each path has had against the threshold.
My audience will never reach 100 per path
Lower the threshold — the floor is 20 — and treat the result as a hint rather than proof. Or use Divide it and read the numbers yourself.
Traffic is all going one way but the weights say 50/50
The test has decided. The summary switches to naming the winner when that happens, and the winning branch is starred.
FAQs
What happens on a tie?+
Path A wins. Something has to, and a coin flip would be worse.
Can I test more than two paths?+
Two per split. You can chain a second split onto a branch, though with each additional path you need proportionally more traffic before any result means anything.
Is this the same as an A/B subject-line test on a campaign?+
Related but not the same. A campaign test compares two subject lines on one send and decides on a timer. This compares two whole paths through a flow and decides on sample size. They’re separate features because the trade-offs genuinely differ.