Skip to main content

How to test AI Inbox

Lucas Kling avatar
Written by Lucas Kling
Updated yesterday

Overview

Testing AI Inbox helps to ensure that it responds correctly to customer requests and follows your studio’s policies.

The best way to test is by creating a few customer test profiles, each set up to represent a different scenario. This keeps the AI focused on the right context and avoids confusion.

In this article, you’ll learn:

  • How to set up test profiles

  • Example test scenarios

  • Best practices for reliable, accurate testing

Step 1: Create test customer profiles

We recommend setting up six profiles, each tied to a different use case.

Profile name

Use case

Emma Info

Active membership with general membership/package questions

Jordan Freeze

Active member eligible for a membership hold/freeze

Taylor Cancel

Active member within cancellation window

Alex Refund

Drop-in or package purchase with refund questions

Mia Schedule

Has active class/appointment bookings

Chris Waitlist

On a waitlist for a full class

Sam Newbie

Prospective member*

*Must have a payment method on file to book. Soon, AI Inbox will be able to send a secure, one-time link so new customers can easily add their payment details.

Note: Keep each scenario separate. Mixing them in one test account can confuse the AI.

Step 2: Run test scenarios

1. Membership information (Emma Info)

Example questions:

  • “What’s included in my membership?”

  • “When does my package expire?”

Expected AI behavior:

  • Pulls details from your membership/package descriptions.

  • Explains clearly or escalates if details are missing.

2. Membership freeze (Jordan Freeze)

Example request:

  • “Can I freeze my membership for two weeks?”

Expected AI behavior:

  • References your freeze policy (eligibility, duration, fees).

  • Offers next steps or escalates if outside policy.

3. Membership cancellation (Taylor Cancel)

Example request:

  • “I’d like to cancel my membership.”

Expected AI behavior:

  • Provides cancellation process and rules (notice period, fees).

  • Escalates if manual review or approval is needed.

4. Refund request (Alex Refund)

Example request:

  • “I was charged for a class I couldn’t attend. Can I get a refund?”

Expected AI behavior:

  • Refers to your refund/credit policy.

  • Processes if eligible, or escalates otherwise.

5. Schedule inquiry (Mia Schedule)

Example questions:

  • “What classes do I have this week?”

  • “When is my next appointment?”

Expected AI behavior:

  • Looks up upcoming bookings.

  • Confirms date, time, and location.

6. Booking or modifying a class (Mia Schedule)

Example requests:

  • “Book me into Pilates on Tuesday at 6 PM.”

  • “Cancel my Thursday class.”

  • “Switch my Saturday appointment to Monday.”

Expected AI behavior:

  • Books, cancels, or reschedules if allowed.

  • Explains late-cancel/no-show rules when relevant.

  • Escalates if host approval is required.

7. Waitlist management (Chris Waitlist)

Example requests:

  • “Can you put me on the waitlist for Friday’s class?”

  • “I heard a spot opened, can you confirm?”

Expected AI behavior:

  • Adds to waitlist if available.

  • Confirms auto-enrollment or waitlist status.

8. Intro offer for new customers (Sam Newbie — optional)

Example request:

  • “I’m new. Can I sign up for an intro offer?”

Expected AI behavior:

  • Recommends intro pass or membership.

  • Books offer only if a payment method is on file.

  • Escalates if no payment method is available.

Best practices for testing

  1. Use natural, conversational language. Example: “Can I freeze my membership?” (not “TEST FREEZE”).

  2. Test one scenario per profile. Keeps context clear for the AI.

  3. Ensure membership and class descriptions are accurate. The AI pulls directly from your setup.

  4. Treat escalations as success. Proper escalation means that the AI is working correctly.

  5. Avoid constant re-training. Only adjust custom instructions when necessary.

Related articles

Did this answer your question?