56.1 Purpose of A/B Testing

AIUNIFY Chat provides A/B testing for evaluating different Agent variations against portions of visitor traffic.

The interface describes this as testing Persona, CTA, or Trigger variations with sticky assignment per visitor.

56.2 Opening Experiments

Open the Agent and select Experiments.

The page heading is A/B testing.

56.3 Experiment Kinds

The Create Experiment form provides three experiment Kinds:

• Persona
• CTA
• Trigger

Choose the Kind that corresponds to the behavior you intend to compare.

56.4 Naming the Experiment

Enter a clear Name that identifies what is being tested.

For example, a CTA experiment might use a name such as “CTA copy test.”

56.5 Variants

A new experiment starts with two variants:

• control
• treatment

Both begin with a weight of 50 in the default form.

56.6 Traffic Weights

Each Variant has a numeric Weight.

The interface explains that weights are relative traffic shares and recommends keeping the distribution reasonably balanced when you want readable comparisons.

56.7 Adding More Variants

Select Add variant to create another Variant.

The deployed form allows up to five Variants and requires at least two to remain.

56.8 Creating the Experiment

1. Enter the Experiment Name.
2. Select Persona, CTA, or Trigger.
3. Name the Variants.
4. Set their relative traffic Weights.
5. Select Create experiment.

56.9 Experiment Statuses

Experiments can display these statuses:

• draft
• running
• stopped

56.10 Starting an Experiment

For an experiment that is not currently Running, select Start.

Starting moves the experiment into Running state and begins its applicable visitor assignment.

56.11 Stopping an Experiment

For a Running experiment, select Stop.

The experiment moves to Stopped state and records the stop point.

56.12 Existing Experiment Summary

The Existing Experiments list displays:

• Experiment Name.
• Status.
• Kind.
• Variant count.
• Variant names.
• Variant weights.

56.13 Deleting an Experiment

Select the delete control and confirm Delete this experiment?.

Before deleting a completed experiment, preserve any outcome information your team needs for future decisions.

56.14 Experiment Best Practices

• Test one meaningful concept at a time.
• Use understandable control and treatment definitions.
• Avoid extreme traffic imbalance unless intentional.
• Do not change the meaning of a Variant midway through a live test.
• Stop the experiment before making conclusions from incomplete traffic.