Put your tricky cases to the test
A customer in a hurry, a vague request, missing information: prepare for these situations before you meet them live.
AI agent testing:
put your agent through its paces before your customers do
Describe a customer, set your criteria and review your agent’s answers. You fix its instructions before the next real conversation.
Available in France today. Coming soon to the UK and Europe.
Test run · phone reception2passed1to fix3tests
Say it is an AIPassed
Find out what the caller needsPassed
Stick to available slotsTo fix
“I’m only free on Tuesday afternoon.”
Testing before your customersA simulated customer.
A simulated customer.
Your criteria, checked one by one.
The simulated customer talks with the agent for up to 30 exchanges. You describe their profile, what they want and what counts as success.
Agent testing · examples to explore
The customer to simulate
Appointment booking
A customer in a hurry, only free on Tuesday afternoon.
2 passed, 1 failedout of 3 criteria
- Say it is an AIPassed
- Ask about availabilityPassed
- Respect the requested slotFailed
Excerpt of the simulated conversation
Simulated customerI can only come on Tuesday afternoon.
LéaI can offer you Tuesday at 10:00.
The instruction to clarifySpecify that only slots matching the stated availability should be offered.
Look at the answer and the last criterion.
Calendar · Tuesday 10:00 and 14:00 available. Simulated result, no tool actually run.
Before every go-liveSpot what needs to change
Spot what needs to change
before the next real call.
Retest after every change
A new instruction can change an answer that used to work. Rerun your cases to see what still passes.
Keep your tests ready to rerun
Sort them into folders and run them in batches. Your team reuses the same cases from one check to the next.
Your scenarios, your standardsDescribe the customer.
Describe the customer.
Write down what you expect.
You write your criteria in plain language: understand the request, respect someone’s availability, don’t make up information.
The simulation can pick up from a conversation opening you have written, and use the contact’s variables, such as their first name.
The classification test starts from a given conversation: it checks the label assigned, such as “Appointment” or “Not interested”.
See how conversations are classifiedA simulation scenario
- Profile
- A customer in a hurry
- Intent
- Book an appointment
- Constraint
- Tuesday afternoon only
- Variable
- Contact’s first name
- Criterion
- Only offer a suitable slot
Conversation opening: “Hello, I’d like to book an appointment.”
Tools stay inside the simulation
Check the calendarSimulated tool
- Set result
- Tuesday, 14:00 available
- Set delay
- 2 seconds · example value
- Appointment created
- None
Sub-agents can be simulated too.
A test environmentTest the answer.
Test the answer.
Without touching your real tools.
You set the result and the delay of simulated tools and sub-agents. The simulation creates no appointments and no records.
Call transfer can’t be simulated. Check this path with a real test call, with your team.
Go beyond the simulationListen, chat,
Listen, chat,
then adjust.
Automated tests and real trials give you complementary views.
A real call to your number
The editor’s “Test call” has your agent call your phone. Listen to its voice and check how the call goes.
See the Phone channelA written conversation in the chat
The test chat lets you talk with your text agents, ask your questions and check their answers.
See the chatbotThe specialist in its test area
Check the sub-agent’s behaviour on its own, then how it steps in during a full conversation.
See sub-agentsVersion historyA change to review?
A change to review?
Go back to the previous version.
Each version keeps your agent’s name, instructions and tools, with its date. You can go back to an earlier version.
After a change, rerun your tests. If an answer no longer fits, compare the versions and restore the previous settings if needed.
Agent history
LéaReception agent versions26 October · 09:30Reception · new instruction
Name · instructions · tools kept
Current version25 October · 17:00Reception · previous version
The settings from before the change
Available to restoreYour setupCheck your agents after every change.
Check your agents after every change.
Start with a trial.
- 01
Describe the customer to simulate
Give their profile, what they want and the constraints of their request.
- 02
Write your success criteria
Say what the agent must do, ask or avoid during the conversation.
- 03
Run, fix, rerun
Read the criteria and the full conversation. Adjust the instructions, then rerun your cases.
Start with peace of mindYour questions,
Another question?Write to us
Your questions,
straight answers.
In short. Test your Starleads agent before you trust it with your customers’ requests. A simulated customer talks with it for up to 30 exchanges, and each of your success criteria is checked separately.
Another question?Write to us Simulation has a fictional customer talk with the agent and checks your criteria. Classification checks the label assigned to a given conversation, for example “Appointment” or “Not interested”.
No. Tools and sub-agents are simulated, with a result and a delay that you set. The simulation creates no appointments and no records.
No. Call transfer can’t be simulated. Check this path with a real test call, with your team.
The number of tests passed and failed, the detail of each criterion and the full conversation. You can go back to the conversation to understand a result.
Yes. You can sort tests into folders and run them in batches, for example after changing the instructions.
Use the editor’s Test call to receive a call on your number. For text agents, use the test chat. Each sub-agent also has its own test area.
Starleads plans start at €79 excl. VAT a month with a 12-month commitment. The pricing page shows which features each plan includes, so you can see which one suits your project.
Starleads is coming to the UK and Europe.
Starleads is available in France today. Join the waitlist and we’ll email you as soon as it opens in your country.