Configuring A/B Testing
Updated
A/B Testing allows you to experiment with different chat experiences and measure what resonates most with your customers. Experiments are created for individual Live Chat applications, meaning each application can have its own set of tests tailored to its audience and use case.By running controlled experiments, you can validate hypotheses and make confident, data‑driven decisions.
Note: This feature is in Limited Availability (LA). To enable this feature in your environment, contact Sprinklr Support at tickets@sprinklr.com.
This guide walks you through the process step by step:
Open A/B Testing
Configure Test Details
Define Target Audience
Configure Variants
Set Traffic Percentage
Schedule the Experiment
Start and Stop the Experiment
1. Open A/B Testing
To open A/B Testing, follow these steps:
1. Click the New Page (+) icon to open the Launchpad.
2. Under Sprinklr Service > Listen, click Live Chat Care.
3. On the Live Chat Applications screen, locate the application for which you want to create an experiment.
4. Click the three-dot menu alongside the Live Chat application.
5. From the menu, click A/B Testing.
6. In the A/B Testing screen, click Create New (top‑right).
The experiment configuration screen opens with two sections:
Test Details
Configure Variants
2. Configure Test Details
In the Test Details screen, configure the following fields:
Field | Description |
Name | Enter an experiment name. For example, FR Greeting Test. |
Description | Enter a description that defines the hypothesis being tested. For example, "Test whether a localized French greeting improves chat initiation rates." |
Start Date | Select the experiment’s start date. For more information, see the Scheduling section. |
End Date | Select the experiment’s end date. For more information, see the Scheduling section. |
Scheduling controls when the experiment goes live and when it auto-ends. An experiment can be started immediately, or saved with a future start time and end time.
Note: Once an experiment is running, only the end time can be edited; everything else is locked.
When the configured end time is reached, a Running experiment moves to Ended automatically.
3. Target Audience
Audience targeting controls which users are eligible for this experiment. Only users matching the configured conditions will be a part of the experiment and get assigned a variant randomly. Everyone else continues to see the default Live Chat application.
From the Target Audience dropdown, select one of the following:
All Visitors
Custom Audience
For a Custom Audience, configure the conditions that determine which users are eligible to enter the experiment. You can apply one or more conditions and select the appropriate operators and values.
Supported Conditions
Category | Condition | Description |
User Browsing Info | Current Page Url | Target users based on the page they are viewing. |
Country | Target users by the selected country. | |
City | Target users by the selected city. | |
Region | Target users by the selected region. | |
Is user authenticated | Include only logged‑in users. | |
User Info | User Device | Target by device type (mobile, desktop). |
Click Next in the bottom-right corner to go to the Configure Variants screen.
4. Configure Variants
The variant step is where you define the actual experience differences you want to compare. The variant editor intentionally mirrors the default Chat Application builder, so there is no new interface to learn the same controls for welcome message, home screen, conversation screen, theme and cards are available inside each variant.
When you open the Configure Variants, the first section shows the Control Version. It is a snapshot of the current default Chat Application at the time the experiment is created. It serves as the baseline experience against which all test variants are compared.
Note: If the default chat application changes later, the Control Version does not automatically update unless explicitly synced after making changes on default app.
Steps to Create a Variant
To create a variant, follow these steps:
1. Scroll to the bottom or minimize the Control Version.
2. Click +Variant.
3. In the Add New Variant dialog box, enter a descriptive name for the variant.
4. Click Add.
The new draft variant is added that you can edit further. Each time a new variant is added, it automatically reflects the complete configuration of the current control variant. Use this copied configuration as the starting point, then change only the specific fields you want to test in that variant.
Surface | What you can change | Typical use |
Theme | App and trigger theme | Test visual appeal of the entire Live Chat |
Home screen | Title, description, brand logo, cards | Test layout, wording changes, and self‑service prominence |
Conversation screen | Title, description, input placeholder, welcome messages | Test in‑conversation clarity |
Note: For Config Filters and any theme‑level changes not supported in the builder, contact Sprinklr Support at tickets@sprinklr.com.
Unsupported Variant Changes
What cannot be changed inside a variant:
Conv disclaimer config, trigger icon conditions, header and chat actions
Chat composer actions: Attachments, speech-to-text, location share, character limit
Video calls and Co-browsing
Advanced settings: one open conversation per user, persistent menu, attachment types, sound notification, sneak peek, chat labels by locale
Note: This list may not cover every unsupported feature. If you need to confirm whether a specific feature is supported, contact Sprinklr Support at tickets@sprinklr.com for more details.
Best practices
Test one meaningful change per experiment to keep the learning clean.
Make the variants visually and behaviourally distinct enough to drive a real difference.
Use the variant name to describe the change, not just 'A' / 'B', it makes reporting much easier to read later.
5. Traffic Percentage
After creating a variant, assign the Traffic Percentage between the Control Version and all variants. The combined allocation must equal 100%.
Common Splits
Split | When to use | Trade‑off |
50 / 50 | Two‑variant tests where both versions are equally trusted | Most statistically efficient |
70 / 30 | When the new variant is riskier or unproven | More traffic needed to reach the same statistical power |
33 / 33 / 34 | Three‑variant tests where all versions are equally trusted | Each variant accumulates data more slowly; plan for a longer run |
Click Create at the bottom-right corner.
6. Scheduling
Scheduling controls when the experiment goes live and when it auto-ends. An experiment can be started immediately, or saved with a future start time and end time.
Steps
1. In the variant wizard, open the Test details page.
2. Select a Start time as a specific date and time in the future.
3. Select an End time. This is when the experiment automatically ends.
4. Click Save.
Note: Once an experiment is running, only the end time can be edited; everything else is locked.
7. Start and Stop the Experiment
After configuration, the experiment moves through a defined lifecycle. Use Start to begin assignment, Stop to end an active experiment manually. To read about the entier experiement lifecycle - (hyperlink the document))
Start an Experiment
If you have not scheduled the experiment, you can click on three dot menu against experiment and click on start.
Once confirmed this will automatically start the experiment instantenously. When the configured end time is reached, a Running experiment moves to Ended automatically.
Stop an Experiment
You can manually end an experiment at any time. For example, if it was set up incorrectly or if the minimum sample size has been reached. Once ended, the experiment stops immediately for all users. Once an experiment is ended, there is no way to restart the same experiment.
You can click on three dot menu against experiment and click on End. Once confimred, this will end the experiment instantenoulsy.
