Quick start: Make your first outbound call

Workflow

The end-to-end process for building, validating, and running an outbound call with a communication agent follows this flow:

1. Enter communication agent management.

2. Create an agent from a template.

3. Build the agent.

4. Debug online.

5. Run test cases for regression.

6. Publish an available version.

7. Run a small-sample calling task.

8. Analyze task results.

9. Extract problematic test cases, iterate, and return to Step 3 to refine the agent.

This guide covers Steps 1 through 7 in detail. Steps 8 and 9 form an ongoing iteration loop that you repeat after each round of small-sample validation. Creating an agent from a template takes about 5 minutes. The full workflow, including debugging, testing, publishing, and small-sample validation, typically takes 30 to 60 minutes depending on agent complexity.

Prerequisites

• You have access to Intelligent Contact Center (AICCS) in the Alibaba Cloud console.

• You have the required permissions to create and manage communication agents. If you cannot see communication agent management in the navigation pane, contact your account administrator to grant the necessary RAM permissions.

• You have an internal test phone number available for call testing. Use a number that your team controls so that you can answer test calls and verify audio quality.

• You have a caller number (instance number) for outbound calling. If you do not have one, apply for an instance number before you reach Step 6.

• You have at least one pre-approved voice qualification issued by Voice Service (VS). Publishing and outbound calling require an approved qualification. If you do not have one, submit a qualification application in Voice Service (VS). For details, see Submit an enterprise qualification.

Step 1: Enter agent management

1. In the left-side navigation pane, click Intelligent Contact Center.

2. Expand communication agent.

3. Click communication agent management.

4. To find an existing agent, search by Agent Name or Agent Code.

Result: The agent list page appears with a Create from Template entry. If the button does not appear, verify that your account has the required permissions.

Step 2: Create an agent from a template

1. Click Create from Template.

2. Search by template name or filter by scenario category.

3. Click a template card and review the applicable scenarios and mode shown on the right panel.

4. Check the Prompt Configuration or Dialog Flow to confirm the template aligns with your customer's business requirements.

5. Click Next.

6. Fill in the agent name, description, brand, and industry. If the scenario is pre-filled from the template, do not change it to an unrelated business type.

7. Click Confirm Creation.

Result: The agent details page opens. The agent is in draft status and ready for debugging.

Step 3: Debug the agent

1. On the agent details page, click Debug in the upper-right corner to open the Online Debug panel.

2. If the agent uses variables, expand Dynamic Variables and enter test values for required fields.

3. Switch to Text Test. Enter a typical customer response to verify that the agent replies as expected.

4. Switch to Call Test, add an internal test phone number, and click Dial Test to validate the real call experience.

Result: The agent responds correctly in both text and call tests. The voice output sounds natural, and the conversation flow matches the expected business logic. If the agent responds incorrectly, adjust the prompt or dialog flow and retest.

Step 4: Run regression test cases

1. On the agent details page, switch to the Test tab.

2. Create or associate test cases that cover the following scenarios:

   • Main flow: The customer follows the expected conversation path from greeting to completion.

   • Rejection: The customer explicitly declines or asks to stop the call.

   • Busy or no answer: The call is not answered or the line is busy.

   • Abnormal input: The customer provides unexpected, off-topic, or silent responses.

3. Select the test cases and click Batch Run.

4. In Run Records, review failed items. After modifying the draft, rerun the failed scenarios and the associated regression suite.

Result: All test cases pass in Run Records. A test case passes when the agent's actual response matches the expected response defined in the test case. If any test case fails, return to the agent draft, fix the issue, and rerun the full regression suite.

Step 5: Publish the agent

1. After tests pass, review the current branch, draft, and version status.

2. Click Publish. Select a pre-approved voice qualification from the list and submit the publishing request. Only qualifications that have already been approved are available for selection.

3. On the version page, verify that the approval status shows Approved. Outbound calls are only allowed after the qualification is approved.

Result: The version page shows the new version with an Approved status. If the status remains pending, wait for the approver to complete the review.

Step 6: Create a calling task

1. Navigate to communication agent > Calling Task Management.

2. Create a calling task and associate it with the published agent version.

3. If you do not have a caller number (instance number), Create a service instance before proceeding.

4. Import a small-sample contact list or use an internal phone number. Do not import a full production contact list at this stage.

Result: The calling task is created and the contact list is imported. The task status shows ready or running.

Step 7: Validate with small-sample calls

1. Start the calling task with the small-sample contact list.

2. After the calls complete, verify the following:

   • Call result statuses: Confirm that each call shows the expected result status, such as completed, no answer, or busy.

   • Tag callback triggers: Confirm that tags are correctly applied and callbacks fire based on the conversation outcome, such as customer intent or rejection.

   • Audio quality: Listen to call recordings to verify that the voice output is clear and the conversation flow is natural.

3. Only expand the contact list scope after the results meet expectations.

Result: Call results and tag callbacks match expectations. The agent is validated for production use.

Analyze results and iterate

After each round of small-sample validation, review the task results to identify problematic conversations:

• Extract failed or unexpected conversation paths from the call records.

• Add these paths as new test cases in the regression suite.

• Return to Step 3 to refine the agent's prompt or dialog flow, then repeat Steps 4 through 7.

Troubleshooting