A subject-line A/B test changes only the subject. Full versions let you change the whole email: subject and body together. You write up to eight versions, a slice of your audience receives them split evenly, and the rest of the list is held back. When you promote a winner, the held recipients receive that version. The section lives on the A/B test step of the campaign composer, under the subject-line test, and it is described there as: send different subject lines and bodies to a slice of your audience, then send the winner to everyone else.

1. Write the campaign’s subject and body on the Content step, and save the campaign as a draft.
2. Open the A/B test step and find Full versions.
3. Select Add full versions. That creates Version A and Version B, both copies of your campaign copy.
4. Edit Version B into the challenger, then select Save version B. Each version has its own Subject field (up to 200 characters) and its own Body editor.
5. To test more than two, select Add version C, and so on up to Version H.
6. Set the test settings below the versions, then send or schedule the campaign as usual.

Each version row has a Pause button (which reads Resume once paused) and a Remove button. A paused version keeps its copy but stops taking a share of the test group, and the word Paused appears beside its heading. Removing asks you to confirm first: Remove this version? warns that the version and its copy are removed from the campaign and that it cannot be undone, with Remove version and Keep version as the choices. Saving each version is a separate action, so editing one version never discards what you have typed into another.

Before you can start, the section explains anything standing in the way:

• The campaign has never been saved: Save this campaign as a draft first. Versions attach to a saved campaign.
• The campaign has no subject or body yet: the start button is disabled and the section says to write the campaign subject and body first, because each version starts as a copy of them.
• A subject-line test is already on: A campaign runs one test at a time. Turn off the subject line test above to write full versions instead.
• The versions could not be loaded: the section says so, confirms nothing has changed on the campaign, and offers Try again.
• Eight versions already exist: 8 versions is the limit. Beyond that the test group is split too thin to tell the versions apart.

Under the settings, a live line shows the split as you change it: how many recipients are in the test group, how many that is per version, and how many are held for the winner. If the split leaves each version with a small audience, a quiet sentence appears beside it. Below about 250 recipients per version it says that 250 per version is where an early signal becomes usable and 1,000 is where a reply rate difference is worth acting on, and that differences below that are usually noise. Between 250 and 1,000 it says the sample is enough for an early signal but not yet worth acting on. If the test group is too small for the campaign to decide at all, a second line explains that a winner can only be decided once every version has at least 30 delivered sends, and gives the smallest test group your number of versions needs. None of these lines block anything: they are advice, and you can send anyway.

Open the campaign and look for Full version test. A chip beside the heading reads Running or Winner promoted, and a summary line underneath states the deciding metric, the test group percentage, the test length in hours, and the date the clock started from. If the campaign extended the window once, the line ends with a note that the window was extended once.

The table has one row per version, with these columns:

• Version: the letter, its subject line, and the word Winner or Paused where it applies.
• Sent, Opened, Clicked, Replied: the counts recorded for that version. On a plain-text campaign the Opened cell reads Unavailable rather than a zero, because there is no open pixel to fire.
• Sample: reads Enough to compare or Still collecting. It tells you whether a version has enough delivered sends to take part in the comparison, not whether the comparison will produce a winner.
• Winner: a Promote button on each version. This column appears only while the test is running and only for people who can edit campaigns. A paused version cannot be promoted.

Under the table, a line states which version is ahead on the deciding metric and whether that lead is larger than what chance produces at the sample you have so far. A winner is decided once every version has at least 30 delivered sends, the test has gathered enough of the deciding signal in total, and one version is ahead by more than chance. The signal is counted in the deciding metric itself: replies for a reply-rate test, opens for an open-rate test, clicks for a click-rate test.

Promoting is a decision you make. If nobody promotes a winner before the test window closes, the held recipients are released automatically and you are told which version was chosen and why. If the automatic check is unavailable, the footnote says so and asks you to promote a version yourself instead of promising a release that will not happen.

Selecting Promote asks Promote this version? and explains that the version is sent to every recipient held back from the test, that the other versions stop sending, and that it cannot be undone. The buttons are Promote version and Keep testing. After a winner is promoted the composer stops accepting edits to the versions and says a winner was promoted and only the winning version sends now.

Promoting by hand is the main path, but the held recipients are never left waiting forever. When the test length has passed, one of these notifications arrives:

• Variant test needs more data: window extended. The test has not gathered enough delivered sends, or enough of the signal it is being decided on, to pick a confident winner, so the window is extended once. The notification says by how many hours, and warns that if there is still not enough data afterwards the first version goes to the rest of the list.
• Variant test: not enough data, kept variant A. Even after the extension the test did not reach the floor, so the first version was sent to the remaining recipients.
• Variant test: no clear winner. There was enough data, and every version performed within the range chance produces at that sample size. Nobody won, so the first version went to the rest of the list. This one is telling you something different from the message above it: the answer is to write versions that differ more sharply, or to test on a larger group, rather than simply to send to more people.
• Variant X promoted. A winner was chosen, either by you or by the test, and it is being sent to the rest of the list.
• Variant test did not start. The test was cancelled at launch, so everyone received the main version and nobody was held back. The message says why: fewer than two versions were turned on, there were not enough recipients to give every version a fair share, or the automatic winner check was not available at that moment. It also says how to run the test again.

These appear in the app. They are listed as Variant test decision in your notification settings, where you can turn them off, or turn email on for them. When email is on they arrive in a digest rather than as an immediate message.

• Versions can be written and edited while the campaign is still a draft. Once the campaign is sending or a winner has been promoted, the copy is locked.
• At least two versions must be active to send. If you pause or remove your way down to one, Review lists the blocker Keep at least two versions active, or remove the full versions.
• Duplicating a campaign does not copy its versions or its test settings. The copy is a plain draft, so you author versions on it again.
• Everyone in the test group receives one version, and each version keeps its own subject line. Follow-ups are unaffected by which version a recipient received.