Call task: create, execute, and analyze-Artificial Intelligence Cloud Call Service(AICCS)-阿里云帮助中心

After the agent is debugged and deployed, initiate real calls following the sequence: Prepare -> Create task -> Import contact list -> Small-sample validation -> Scale up -> Review and optimize. Then review execution records and call results.

Quick path

Verify that the communication agent has at least one published and live version, and confirm the current business scenario.
Create a task and configure time slots, concurrency, unanswered-call retry, and trunk line information.
Import a contact list based on the current template, and verify phone numbers and variable fields.
Start with a small sample first; check connection rate, opening sentences, variables, tags, recordings, and the actual version used.
Scale up only after all of the above checks pass; if issues arise, pause the scale-up, locate the problem, and run regression validation.
In the task details, review effectiveness by execution record, call result, branch, and version; capture issues as test cases.

Create and configure a task

Go to LLM communication > Communication agent > Call task management, click Create call task, and check the following:

Configuration item	Key points
Task name	Use a searchable, batch-distinguishable name; avoid temporary names with unclear meanings.
Communication agent	Select the agent to deploy; no need to manually lock the version on the creation page. For the actual version selection logic, see Agent branch and version effective rules.
Task concurrency	Calculation logic: calls per second + calls currently in progress.
Associated service instance	Select the service instance to associate. See Create a service instance.
Start method	Choose immediate or scheduled start based on business needs, and confirm working days, callable time windows, and pause windows.
Unanswered call retry	Only applies to cases where no effective communication was established (unanswered, rejected, etc.); configure retry reason, count, interval, and stop boundary based on business needs.

Note

For scenarios where the customer requests to be called back after the call is connected, use the "Redial" rule in the agent's Post-call actions configuration configuration; do not mix this with the task's unanswered call retry.

Import a contact list

Download the current contact list template from the task and fill in data according to the template fields.
Fill in the called phone numbers and the input variables required by the agent; field names, formats, and null-value handling must match the agent configuration.
After uploading the contact list, verify the import count, failed rows, and field mapping; validate with a small number of real or controlled numbers first.
The pending-call records in the contact list will confirm the actual agent branch and version to use when they enter the scheduling queue. See the next section for details.

Agent branch and version effective rules

Before each call record is actually initiated, the system reads the latest published version of the active branch. A single task may use multiple versions; when a paused task is resumed, unstarted records are executed using the version at the time of resumption. Before publishing a new agent version or switching the active branch, assess the impact on tasks currently running.

Scenario	Branch and version handling
Creating a task or importing phone numbers	Associates only the communication agent; does not pre-lock the version for all records.
Call in progress	Active calls continue using the configuration confirmed at call start; publishing a new version or switching the deployed branch afterward will not affect ongoing calls.
Task paused, record still pending	Pausing does not overwrite pending records. After the task is resumed, when the record re-enters the scheduling queue, the system re-confirms the deployed branch and the latest live version at that time.
Viewing task results	In the Task details page, view the branch and version actually used for each record in the Execution records tab; a single task may show multiple versions.

View task details and execution records

After entering the task details, check the basic task information and overall progress, then view the execution records. Execution records are classified by the following status types:

Record type	Purpose
Completed	View results, tags, duration, conversation details, and the actual branch and version used for completed calls.
Pending	View records that are still waiting for scheduling or have not yet been executed, or are waiting for retry; after the task is resumed, these records will re-confirm the version when they enter the scheduling queue.
Failed	View records that could not be executed and the failure reason; determine whether the issue is a data, trunk line, scheduling, or configuration problem.

After opening an individual Call details record, view the available conversation log, recording, output tags, variables, and call result; use the branch and version recorded in this record as the basis for problem reproduction and root cause analysis. For bulk review, export execution records or recordings using the page's export function.

Quick path

Create and configure a task

Import a contact list

Agent branch and version effective rules

View task details and execution records

Related API documentation