Skip to main content
Belcenter AIHelp center
Outbound calls

Create an outbound campaign

Import a file or supply JSON contacts, trigger a one-off call via the API, test, control pacing, and monitor final webhook delivery.

A campaign calls one or more contacts by providing the bot with shared context and row-specific data. It is also the public API entry point to request a one-off call with a bot.

Prepare the file

Use an Excel or CSV file with a header row. One column must contain the phone number. The other columns can provide the name, a reference, an amount, or any data useful for the scenario.

Include only information necessary for the call in the file. With the API, you can provide the same list directly in the contacts JSON array without creating a file.

Create the campaign

  1. Name the campaign and choose the bot.
  2. Describe the common objective.
  3. Upload the file, then select the phone number column.
  4. Map the other columns to labels that the bot can understand.
  5. You must select the campaign type: direct marketing, marketing with explicit consent, or transactional.
  6. Set the date. To adjust the pacing and maximum number of simultaneous calls, open Delivery settings.
  7. If an external system needs to be notified, open Completion notification and enter its HTTPS URL.
  8. Save as draft or schedule.

Existing delivery and notification settings continue to apply when their panels are closed. Costs, campaign type, and consent requirements remain directly visible. A new campaign created in the interface is a live campaign; no simulation option is offered there.

When the call is answered, the bot always lets the callee speak first: it waits for a greeting such as "hello", "yes", or another utterance before delivering its introduction. If no one speaks for five seconds, the runtime releases the introduction as a fallback to prevent indefinite silence. This technical instruction applies equally to campaigns created in the interface, via the public API, and to test calls.

Test without placing a call

Simulation is restricted to the API on platforms where this feature is enabled. Create it and apply its results using the operations below. The interface does not offer any simulation options or mock result controls. Campaigns created via the API remain identified as simulations in the list and via a notice on their details, edit, and duplicate screens.

No numbers are dialled, no AI providers or business tools are run, no credits or minutes are consumed, and no external DNCM checks are carried out. Results are labelled [SIMULATION]. This feature tests transport and statuses, not voice quality or bot behaviour. It does not replace business testing on mock records.

A simulation can never be converted into a live campaign. Editing or duplicating it, including in the interface, automatically preserves this mode; do not supply an simulation_mode field in a form. An outdated or tampered form that submits this field is rejected without creating or modifying any campaign. The legacy Test call button, which actually places a phone call, cannot be used on a simulation.

Permissions remain identical to campaigns: you must have permission to edit the organisation's bot. A platform without this feature will reject the request without falling back to a live call.

The API uses a separate endpoint: POST /api/v1/outbound-campaigns/simulations, with the same body as standard creation, notably schedule: true. Verify data.simulation_mode: true in the response. Never retry via the live creation route if this route is unavailable.

To apply a result:

POST /api/v1/outbound-campaigns/123/calls/456/simulate
Authorization: Bearer <clé-api>
Content-Type: application/json

{"outcome":"answered","summary":"Le client promet de régler la facture, qui reste impayée."}

Possible values are dialing, answered, no_answer, busy, failed, and voicemail. Repeating the same result is idempotent; overwriting a terminal result is rejected with HTTP 409. When all calls are finished, the standard webhook is triggered, complete with delivery retries and recovery. This webhook may modify your connected CRM: use only a test database and test records. The summary remains fictitious and does not certify any business action.

In Odoo, the simulation setting is available with BeAI Development. It only flags new calls; live calls already pending are paused. After sending a simulation, Simulate an answer / Simuler une réponse allows you to test the returned result. Other scenarios are available via the API. Disabling this setting does not turn past simulations into live calls.

Trigger a one-off call via API

The public API does not provide a direct /api/v1/outbound-calls route. To call a single contact, create a campaign containing a single entry and schedule it in the same request:

POST /api/v1/outbound-campaigns
Authorization: Bearer <clé-api>
Content-Type: application/json
{
  "name": "Appel ponctuel CRM-12345",
  "bot_id": 42,
  "prompt": "Contacter le client au sujet du dossier CRM-12345.",
  "campaign_type": "transactional",
  "schedule": true,
  "max_concurrent_calls": 1,
  "contacts": [
    {
      "phone": "+32470123456",
      "name": "Marie Dupont",
      "case_id": "CRM-12345"
    }
  ]
}

With schedule: true, omitting scheduled_at means launching as soon as capacity, credits, and applicable checks allow. To schedule the call, provide a future ISO 8601 date, for example 2026-09-03T09:30:00+02:00. With schedule: false, the campaign remains in draft and is launched later with POST /api/v1/outbound-campaigns/{id}/start.

The 201 response contains data.id for the campaign and data.calls[].id for each call. Check individual status with GET /api/v1/outbound-campaigns/calls/{call_id}. Terminal states are completed, failed, and cancelled; scheduled, claimed, and dialing indicate that processing is still ongoing.

Before dialling the number, the runtime also checks that the initial prompt, context, and tools fit within the safe budget of the AI model. If this budget is exceeded, no SIP call is placed: the call moves directly to failed, with failure_code: "ai_prompt_too_large" and a readable explanation in last_error. The same information appears in the campaign within the interface and in the API tracking responses.

Choose transactional for a call related to an existing relationship or transaction. The marketing_direct type applies the DNCM check. The marketing_consent type requires consent_evidence_acknowledged: true and assumes the organisation retains proof of explicit consent.

Receive the completion notification

The webhook is optional. BeAI calls it in the background once all contacts in the campaign have reached a terminal state. Temporary downtime triggers multiple retries.

A completed campaign may still have its notification pending: call completion and webhook delivery are two separate stages. In campaign tracking and under GET /api/v1/outbound-campaigns/{id}, check completion_webhook.status, attempts, last_status_code, and delivered_at. A pending status with zero attempts means sending has not yet started; it does not indicate a recipient outage. delivered and an HTTP 2xx response confirm acceptance by the recipient. In the event of an extended delay, contact support without relaunching the campaign, as this could redial contacts.

The same campaign read permissions and organisation scope apply to both the interface and the API. An interruption in background processing may delay a notification; its resumption preserves the delivery ID. The recipient must handle this ID idempotently, particularly if its HTTP response is lost.

The notification remains intentionally minimal: it indicates the outbound_campaign.completed event, the campaign ID, the completed status, a delivery ID, and the GET /api/v1/outbound-campaigns/{id} route. It contains no contact details, transcriptions, or detailed results. The recipient system can use its own BeAI API key to retrieve the information it is authorised to access.

Make a test call

Use Test call to enter your number and representative values for each column. This test checks the prompt and behaviour without launching the entire list.

The API equivalent is POST /api/v1/outbound-campaigns/{id}/test-call. It therefore requires an existing campaign, reuses its bot, prompt, mapping, and type, and returns a call in the same format as the tracking endpoint. The test call remains subject to applicable checks and costs, but it is excluded from campaign counters and progress.

Control call pacing

The selected number remains subject to the overall server limit and your SIP carrier limits. A campaign configured for three concurrent calls may progress more slowly if other calls are already using available capacity.

DNCM check

For direct marketing, BeAI checks the number against the DNCM list. A recent cached decision is used first, followed by the official API if necessary. In the event of downtime without a local decision, the call may proceed in degraded mode and this decision is logged.

Track and reuse

From the campaign, open the pre-filtered conversations. You can delete or edit authorised contacts, duplicate a campaign to reuse its context with a new file, and relaunch a campaign according to the available actions.

If the Recording skill is active on the bot, outbound calls are recorded just like inbound calls. The audio file appears in the conversation details once the call has finished.

Questions during a follow-up call

Prioritising the callee's opening words applies only to the call greeting. The rest of the call is a dialogue: the campaign mission sets the initial objective, without requiring the bot to hang up as soon as a status is obtained. Any follow-up question, such as the date of an engineer visit during an invoice reminder, must receive a verified answer or a clear explanation of any missing information before ending the call. Adding a note or promising human follow-up does not replace this answer.

Documentation updated with the application