# A/B test a sequence

Run two variations of a Victoria AI sequence, split new leads between them, read the verdict, and end the test.

An A/B test runs two versions of a campaign's sequence, Variation A and Variation B, on the same leads list. Each new lead is assigned to one variation and stays on it.

## Turn on A/B testing for a campaign

1. Open **Campaigns → your campaign → Sequence**.
2. Select **Enable A/B Testing**. Variation B starts as a copy of A and the editor switches to B.
3. Change B. Use the **A** and **B** buttons to switch between them.

Make the variations differ in angle, not just wording: a different opening, offer or call to action. Two near-identical versions rarely show a difference. If B is empty, **Copy from Variation A** fills it.

The readiness check adds two checks while A/B is on: **Variation B has steps** and **Variation B steps have content**.

## Set the A/B traffic split

**Split** next to the variation buttons sets the share of new leads that get A: **A 50 / B 50** through **A 90 / B 10**. It applies to leads added from then on. Leads already in the campaign keep their variation.

Leads that were in the campaign before you turned A/B on are **Unassigned**. They run Variation A, count in **All**, and count in neither arm.

## Read the A/B verdict

Open **Campaigns → your campaign → Analytics**. With A/B on, the tab shows an **A/B verdict** card and a funnel for each variation, and a filter at the top (**All**, **A**, **B**, **Unassigned**) narrows every number on the page to one arm.

The verdict compares each arm's positive reply rate (leads with a positive reply ÷ leads contacted) with a two-proportion z-test:

| Verdict | Meaning |
| - | - |
| **Not enough data yet** | An arm has fewer than 50 contacted leads. |
| **No meaningful difference** | The arms are too close to call. |
| **Leaning A** or **Leaning B** | One arm is ahead, with p below 0.2. Keep it running. |
| **A wins** or **B wins** | One arm is ahead, with p below 0.05. |

The arm a lead belongs to is the variation it was assigned when enrolled. A + B + Unassigned always equals All.

## Promote the winning variation

The Sequence tab has no promote button. Ask the AI Sales Copilot, for example "promote variation B on this campaign", and approve the card. A connected assistant can do the same with the `promote_ab_winner` tool.

- Promoting **A** sends every new lead to A. Leads already on B finish B.
- Promoting **B** copies B over A and sends every new lead to it. Leads still in progress on A continue on B's copy from their current step number.

A/B testing stays on after a promotion, so leads mid-way through B finish its copy.

## Turn off A/B testing

Select **Disable A/B**, then **Delete Variation B**. Turning A/B off deletes Variation B's steps. Leads that were on B continue on Variation A from the same step number, which can skip or repeat a step if the two variations have different lengths. To keep B's copy, promote B first instead.