Voice AI testing · Retail
Customers remember whether you followed through. Test order changes, refunds and delivery requests against the records that tell you what actually happened.
Your scenarios. Your success criteria. Evidence to review.
What did the agent promise?
What did the system accept?
What was actually saved?
Illustrative scenarios · fictional conversations and expected checks
Simulated caller
Has the refund for my return been processed?
Agent
A refund of $42 was accepted. Your bank may still need time to post it.
Amount and status matchCompare the promised amount and refund state with the payment record. Test rejected transactions and duplicate retries.
Cancellation not completed
FlaggedEvidence to review
The order was already dispatched. The service rejected the cancellation, but the agent told the shopper it was cancelled.
Test the point where an order can no longer be stopped. A rejected cancellation must not become a spoken confirmation.
Simulated caller
Can you add the side entrance to the delivery notes?
Agent
The delivery note has been saved for this order.
Instruction savedVerify that the new address or instruction is attached to the intended order and that the agent describes its actual status.
Anchor each test to the customer, order and operation that the conversation actually concerns.
Include stock conflicts, return conditions and late delivery changes in the evaluation design.
Keep the failed scenario so the same request can be checked after an agent change.
Fictional commerce examples. Store, fulfilment and payment integrations require a scoped pilot; no purchases, refunds or customer orders are executed by this page.
Bring your team’s hardest scenario. Let’s define what a successful result would need to prove.
Book a demo