How to test an AI receptionist with your own call script
A demo shows the best case. Your clinic has real calls, with real questions and real pauses. A test with your own script shows how the agent handles your work, not the sales team's work.
This guide explains how to build a test script and run it. The test takes a few hours of staff time and gives clear results.
Write the calls your clinic gets
Open your phone log and your voicemail notes. Pick the calls that come up most often. Write each one as a short script. Use the words a caller will use, not the words in your brochure.
Include five common types of call. A new patient asks about a service. A current patient asks to move a visit. A caller asks about a price. A caller asks about a deposit. A caller asks for a provider by name.
Add the hard calls
Easy calls prove little. Add calls that test the limits. A caller can change the subject in the middle of a booking. A caller can give a wrong date and then correct it. A caller can speak fast or ask two questions at once.
Add one call that asks a medical question. The agent must not give medical advice. It must send the caller to a safe path, as your rules say.
Add one call from a caller who is upset. Test whether the agent stays calm and offers a person when the caller asks for one.
Set the test rules
Write the rules before you start. Decide what counts as a pass. A pass means the agent answers the question, stays within your rules and reaches a clear next step.
Decide who makes the test calls. Use two or three people from your team. Different voices and accents show how the agent handles a range of callers.
Run the calls from outside
Call from a personal phone, not from the office line. Call at different times of day. Call at least once during a busy window in your clinic.
Ask a team member to call from a second phone. Two callers at once show whether the agent can handle a second call without a delay.
Score each call
Use a simple sheet with four columns: call type, outcome, notes and pass or fail. Write a short note for each call. Note any answer that sounded wrong or too long.
Look for patterns after the test. Repeated failures point to a missing rule or a missing piece of information. Single failures often show a rare case that needs a small fix.
Check the booking step
After each booking call, open your scheduling system. Check that the visit appears in the right place. Check the service, the provider and the time. Check that the confirmation text went to the caller.
Make a test move and a test cancel. Confirm that the schedule changes and that the caller hears a clear answer.
Review the results with your team
Meet with the team after the test. Review the failed calls first. Ask the front desk staff what they will change in the agent's answers.
Write a list of changes. Decide which changes the vendor makes and which changes your team makes. Retest the failed calls after the changes.
Compare the results with the vendor claims
Compare the test results with the claims in the vendor proposal. Note each claim that the test did not confirm. Ask the vendor to explain the gap in writing. A written answer gives you a clear record for the decision.
Share the test sheet with your team before the final decision. Ask each team member to sign off on the calls they tested.
A worked example
This example is hypothetical. A clinic runs a test with fifteen calls. Most calls pass. Three calls fail because the agent gives a long answer about pricing. The clinic shortens the pricing rule and retests. All three calls pass on the second run.
The test showed a clear problem and a clear fix. The clinic then moves forward with more confidence.
Kepler Voice lets clinics test the agent with their own questions before they decide. The agent, Ivy, answers in a voice that sounds like your clinic. To run your first test call, talk to Ivy at (669) 337-5915.