Understand needs first
Ask about the patient’s interests. Explain the service, then discuss pricing and booking.
AI agent A/B testing
Test your AI conversations against real business results.
Try a change with real customers before making it your main setup. Choose one measurable goal, split eligible traffic between saved versions, and review the evidence with your team.
A clinic example
Does sharing pricing earlier lead to more confirmed bookings?
Eligible new conversations on the selected connected account
Share pricing earlier
Give approved pricing information earlier, then help the patient choose a booking.
Primary goal
Confirmed bookingsEdit B in your existing agent builder. Your main setup keeps running until launch.
Illustrative test design. No customer results are shown. A 50/50 setting does not guarantee equal participant or message counts.
Compare how your agent responds
Keep the original as A. Edit a test copy as B in Agent Flow or Agent Builder. Test a different opening question, instructions or model on a chosen connected account, website widget or API target.
Measure what follow-ups contribute
Run a separate test with automatic AI follow-ups on or off. Customer opt-outs, AI pauses and channel rules still apply. Your staff can continue helping.
Test how website chat appears
Compare launcher-only, welcome bubble and auto-open modes for eligible visitors who consent. A visitor opening the panel does not count as starting a conversation.
In your workspace
Find out which agent changes help your customers take the next step.
- Open A/B tests in Agent Flow or Agent Builder. Choose the test type and connected account, website widget or API target.
- Save your original setup. Edit the test version in the familiar builder and review what changed.
- Choose one primary goal, its measurement window, a fixed traffic split and how the test ends: after a number of days, a participant count, or manually.
- Launch the test. Eligible new participants receive a version; individual messages are not split between versions.
- Review recorded outcomes and the supporting conversations. Allow existing participants time to complete their measurement window.
- Choose what to keep. Apply a tested version or restore the saved original when the current target and dependencies allow it.
What happens next
Test history keeps the saved versions, differences, goal, dates, recorded performance and your decision. You can inspect the evidence later and reuse a saved version after the current setup is checked.
What to know before you start
Available in enabled paid Receptionist workspaces, subject to team permissions. Traffic stays at your chosen fixed ratio. There is no automatic traffic shift or winner promotion. AI usage still uses credits. Missing evidence can prevent a recommendation, and a test can be inconclusive.
Your questions, answered
Which goals can I measure?
Available goals depend on the test type and connected tools. Examples include customer reply thresholds, booking requests, confirmed bookings, captured contact details, completed intakes, business handovers, created orders and paid orders. Pick the outcome before launch; a request is not confirmation.
Does 50/50 mean exactly half the messages go to each version?
No. The configured split applies to eligible participants, not individual messages. Actual counts can differ, and each participant can send a different number of messages.
Can I edit B without changing my live agent?
Yes. B stays in the test draft while you edit it in Agent Flow or Agent Builder. Review and save the changes, then return to setup before launching.
Will the system automatically choose a winner?
No. The traffic ratio stays fixed. You choose which version to use. An optional scheduled end can return to normal routing or use a version you explicitly selected, after target and dependency checks.
Can I review results on my phone?
Yes. Native mobile supports results and permitted controls. Create and edit tests on the web.
Read the practical guides
Connected capabilities
AI conversation controlsGive your AI a clear job and clear boundaries.Cross-channel conversationsKeep useful customer information close to the conversation.Shared team inboxThe assigned teammate can review and continue the conversation.Conversational intake and formsThe right questions, in the right format.