> ## Documentation Index
> Fetch the complete documentation index at: https://www.voiceflow.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Test and launch your integration

> Validate the complete contact-center integration before switching a production route. Exercise the real customer or staff entry point, the receiving workspace and the…

Validate the complete contact-center integration before switching a production route. Exercise the real customer or staff entry point, the receiving workspace and the failure path. A saved connection, prototype Run or accepted API request is not an end-to-end test.

## Before you begin

- A configured [customer connection](/docs/guides/integrations/contact-centers/integrate), [human handoff](/docs/guides/integrations/contact-centers/transfer-calls), or [staff assistant](/docs/guides/integrations/contact-centers/agent-assist).
- A teammate available in the test queue, two fictional cases and test identities with different access.
- Permission to publish the test Environment and change the provider route. Save the previous route, agent configuration and release/version so you can restore them.

<a id="verify-the-customer-journey" />

## Validate and release the integration

### 1. Freeze the test configuration

Record the channel, test entry point, Voiceflow project, Environment, version, receiving destination and expected context fields. Check that the human destination does not send the customer back into the same AI entry route.

#### Studio

Open [Agent studio](https://creator.voiceflow.com), choose the project and test Environment, and publish the reviewed configuration. For a phone connection, verify its Environment assignment and select the intended **Published** version before the release check. For a widget, verify the project and Environment used by the installed test page.

#### Developer

[Publish the test Environment](/docs/api-reference/environment/publish-environment) after checking its saved agent and Workflow configuration. Publishing promotes its draft immediately; use a non-production Environment for this test. Point the customer application, SIP assignment or staff backend at that Environment. Record the resulting release/version with the provider routing configuration.

Keep keys out of customer code and test records. Use correlation identifiers, destinations and HTTP/event outcomes when collecting evidence.

### 2. Exercise the actual entry point

Run each channel you intend to release. Passing one does not validate another.

- **Phone:** call the actual number through the contact-center route. Confirm the correct AI answers, both sides hear speech, a requested handoff rings the staffed destination, and the human and caller can speak without the AI.
- **Chat:** start from the installed customer widget or application. Trigger Handoff, exchange a customer message, human reply and second customer message. Confirm each arrives once and internal staff notes stay private.
- **Staff assistance:** open an authorized fictional case from the staff workspace. Ask for a case-specific suggestion, confirm its context and review it before manually sending anything to the customer. Switch cases and check that context does not carry over.

#### Studio

Inspect the result in the provider workspace and the customer interface, not only Creator Run. Use [Transcripts](/docs/documentation/measure/transcripts) to locate the matching AI conversation. The staff-assistant Run below is a preparation check; the deployed staff application's permissions and case mapping need the separate checks above.

Staff assistant test reply showing policy guidance and a proposed customer response — Check the assistant's response, then test its delivery and permissions in the actual staff workspace.

#### Developer

Capture the receiving system's conversation/call ID and your application's corresponding session or request ID. For Chat, verify the existing Socket.IO session resumes before sending human-support messages. For staff assistance, assert that a denied staff/case request makes no Voiceflow call. Never log the project credential or `sessionKey`.

### 3. Verify context and completion

Check the actual receiving fields against your mapping. A variable name or emitted trace is not evidence that its resolved value arrived at the right team.

- **Phone:** inspect the destination's received caller/context fields and end the call from the human side. Confirm the provider records the expected completion. SIP REFER acceptance and a human answering are separate checks.
- **Chat:** send a final public human reply, then close the provider conversation. Confirm the reply arrives before closure, then verify the configured end or [return-to-AI behavior](/docs/documentation/handoff/return-to-ai) in the same customer session.
- **Staff assistance:** sign in as the staff account without access to the case. Confirm access is denied before runtime execution, and that the normal workspace still works when assistance fails.

### 4. Exercise failure and restore the prior route

Use a separate test configuration to make the downstream connection unavailable. For Chat, confirm initial Handoff creation reaches the configured **Failure path**. For Phone, test the provider's rejected or unanswered transfer behavior; Call forward does not supply an automatic recovery branch. For staff assistance, make the backend or runtime unavailable and verify the staff member can continue normally.

Restore the previous configuration and repeat one successful contact. Verify that no customer is sent into a routing loop and no queued test callback or message reaches a new conversation by mistake. This is the rollback rehearsal.

### 5. Switch the approved route and make a final check

Have the contact-center administrator switch only the route you validated to the approved Voiceflow Environment/version. Repeat one controlled contact through the production entry point with the receiving team available. Verify routing, context, two-way delivery and completion again.

If any check fails, restore the recorded provider route and verified Voiceflow release/configuration. Confirm a fresh contact follows that restored path before reopening traffic. Do not retry an uncertain transfer or message by creating a second customer conversation.

<a id="record-the-result" />

## Completion record

Record the Environment/release, provider destination, observed channel outcomes, acceptance owner and rollback configuration. Release only the paths you exercised.

## Diagnose a failed check

- **Phone entry, audio or REFER:** inspect provider call logs and [SIP troubleshooting](/docs/api-reference/agent-runtime/voice/sip/troubleshooting).
- **Chat routing, replies or closure:** inspect the receiving queue, current session and [Handoff troubleshooting](/docs/documentation/handoff/troubleshooting).
- **Staff case access:** check the authenticated staff principal and server-side case lookup before changing the assistant's prompt.
