How do you A/B test within automated flows?

You've built a welcome flow, a post-purchase sequence, or an abandoned cart series. Now you want to know which subject line, which email timing, or which copy angle actually performs better. That's where A/B testing inside automation comes in, and it works a bit differently from a regular campaign split.

In a standard campaign, you blast both versions at once and get results in 48 hours. In automation, contacts trickle in over days, weeks, or months as they trigger the flow. That changes how you set things up.

Step 1: Add a split at the right point

Most ESPs let you insert a split step directly into your flow. In Klaviyo, it's called a "Conditional Split" or you can use the built-in A/B test block inside the flow. In ActiveCampaign, it's a "Go To" split that routes contacts by random percentage. In HubSpot, workflow branches let you randomize by percentage. Set the split before the email you're testing, not after it.

Step 2: Keep the split simple at first

A 50/50 split is the cleanest starting point. Version A goes down one path, Version B goes down the other. Change only one variable at a time. If you test the subject line AND the send delay at the same time, you won't know which one drove the difference. One variable per test.

Step 3: Know what you're measuring and how long to wait

So this is where automation testing trips people up. Because contacts enter the flow gradually, you need enough contacts in each variation before the results mean anything. A rough rule of thumb: aim for at least 200-300 contacts per variation before you start drawing conclusions. Fewer than that and a handful of outliers can skew the whole result.

Also account for the time factor. A flow triggered by a purchase might take 6-8 weeks to accumulate enough contacts. Don't declare a winner after the first 30 contacts. Let it breathe.

What's actually worth testing in flows

• Subject lines for open rate differences
• Send timing (email 1 sent immediately vs. 1 hour after trigger)
• Number of emails in a sequence (3-email series vs. 5-email series)
• CTA copy or button placement for click-through differences
• Plain text vs. designed emails for reply rate or engagement

One thing to watch for

Long-running tests pick up seasonal noise. A test that starts in October and runs through December will include Black Friday behavior, which skews everything. If your flow runs over a holiday period or a major promotional window, either pause the test or treat those results separately.

Once you have a clear winner, update the flow to send only the winning version. Don't leave two branches running forever just because you forgot to close the test.

Want to check your full automation setup before you start testing? Our SOS hotline is free and genuinely helpful if you're not sure where to start.