> For the complete documentation index, see [llms.txt](https://learning.contextqa.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/booking-outcomes.md).

# Verify Voice AI Bookings and Business Outcomes

Test requested, confirmed, and persisted booking details with explicit rubrics and backend verification so clear speech cannot conceal an incorrect action.

> **Documentation owner:** [Deep Barot](https://contextqa.com/academy/) · **Last updated:** November 11, 2026

## Quick answer

To verify a voice agent’s booking, compare the user’s final request, the agent’s confirmed details, and the persisted backend record. Score date, time, timezone, identity, and authorization requirements explicitly. A clearly spoken confirmation or a high audio score does not prove that the right appointment was created.

## Key takeaways

* Verify final corrected values rather than the first values mentioned.
* Require affirmative confirmation before an irreversible business action when the workflow specifies it.
* Keep conversation accuracy, pronunciation, and backend persistence as separate assertions.

## Define the transaction contract

Before testing, specify the required customer identity, service, date, local time, timezone, availability rule, confirmation rule, and duplicate-prevention behavior. For date-sensitive examples, choose future dates relative to the test environment and avoid ambiguous formats.

An illustrative caller asks for May 30, 2030 at 6:30 p.m. Central time. If the agent clearly reads back a different date, pronunciation may be acceptable while booking accuracy fails. If it confirms the right date but stores the wrong name, the date criterion can pass while identity accuracy still fails.

## Configure separate criteria

| Criterion                     | PASS evidence                                                | Failure example                                  |
| ----------------------------- | ------------------------------------------------------------ | ------------------------------------------------ |
| Final request retained        | Corrected values are used in the final readback              | Agent reverts to an earlier date                 |
| Confirmation before action    | Caller affirms the complete required details before creation | Agent treats “no” as approval                    |
| Identity accuracy             | Required name or customer identifier matches                 | Date is fixed but name remains wrong             |
| Persisted booking matches     | Correlated backend record matches all required fields        | Correct spoken confirmation, wrong stored record |
| Exactly one authorized action | Only the intended booking is created                         | Duplicate or unauthorized booking                |

Create custom metrics using [metric configuration](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/metrics-and-judges.md). Give the rubric atomic pass and fail conditions; decide whether partial credit is appropriate for each requirement. Do not average away a critical wrong-date or unauthorized-action failure.

## Use a verifier for persistence

Configure a read-only REST or Lambda **Verifier** that returns the booking associated with this case. Enable tool grounding on the appropriate metric. Include the fields needed for comparison, and identify the record with a booking or correlation identifier rather than the most recent calendar entry.

Normalize times deliberately: compare the intended local time and timezone with the stored representation, accounting for the date’s timezone rules. A UTC timestamp and a local-time display can describe the same instant, while an omitted timezone can hide a real defect.

Follow [tool verification](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/verify-tool-actions.md) for roles, parameter substitution, and missing-evidence handling.

## Test rejection and recovery

Include cases where the caller rejects a readback, changes the service, requests an unavailable slot, or abandons before confirming. The agent should recover according to the requirements without claiming success or leaving an unauthorized booking.

For concurrency, deliberately prepare two requests for the same slot and specify the expected winner/loser behavior. Do not reuse a single occupied slot across unrelated success tests and call every rejection a defect.

## Evaluate post-call work separately

If the business outcome includes a later CRM update, summary, notification, or follow-up process, define a separate assertion with a correlation identifier and an allowed completion window. The call’s transcript cannot prove that later work happened. Connect the evidence only when the configured workflow actually retrieves it.

Use [run review](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/runs-and-regression.md) to report partial fixes accurately and [voice troubleshooting](https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/troubleshooting.md) when evidence is incomplete.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/booking-outcomes.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
