***

title: Phone Number Transfers (TEL)
subtitle: Transfer calls to standard phone numbers in E.164 format.
slug: transfer-calls-to-humans
description: Learn how to configure call transfers to phone numbers using E.164 format in Synthflow.
---------------------

For clean Markdown of any page, append .md to the page URL. For a complete documentation index, see https://docs.synthflow.ai/llms.txt. For full documentation content, see https://docs.synthflow.ai/llms-full.txt.

Phone number transfers (TEL) let you reuse the same transfer action across different AI agents, even if they use different telephony connections, by redirecting calls to a TEL URI in `tel:<number>` format.

<Note>
  Looking for other transfer types? See the [Call Transfers overview](/call-transfers) for SIP and dynamic transfer options.
</Note>

***

## **Key Settings for Call Transfer**

### **1. Phone Number to Call**

* Specify the phone number that the assistant will transfer the call to. This can be a direct line or an extension within your organization. There is a validation button to confirm that the number is valid.

### **2. When Should the Transfer Take Place?**

* This section defines the condition for the transfer. For example, you can set it to transfer when the user requests to speak to a manager or human agent.

### **3. Background Music**

* Optionally enable background music during the transfer process to ensure a smooth and professional experience for the user while waiting for the transfer to be completed.

***

## Transfer Modes

| Mode               | Description                                        | Best For                                   |
| ------------------ | -------------------------------------------------- | ------------------------------------------ |
| **Blind (Cold)**   | Transfer immediately without speaking to the agent | Simple routing, queues, IVR systems        |
| **Warm (Message)** | Play a private whisper message to the agent        | Providing specific context or instructions |
| **Warm (Summary)** | AI summarizes the conversation for the agent       | Complex calls requiring full context       |

<Note>
  **Whisper messages** are heard only by the receiving agent, not the caller. Use this to brief the agent on the caller's issue, mood, or any relevant account details before they accept the call.
</Note>

## Human Detection (Warm Transfer)

When using a warm transfer (Message or Summary), the AI needs to deliver a briefing to the receiving agent. By default, the briefing plays as soon as the call connects — but that connection might be a voicemail greeting, an IVR menu, or empty air.

**Human Detection** solves this by waiting for a live human to answer before delivering the whisper message or summary. This ensures the briefing is heard by a real person, not wasted on an automated system.

### Human Detection Timeout

Set how long the AI should wait for a human to be detected before giving up. If no human is detected within this window, the transfer is treated as failed and your configured failure handling applies. Minimum 30 seconds, maximum 30 minutes.

<Note>
  Human Detection is only available for **warm transfers** (Message and Summary modes). Blind transfers connect instantly and do not support detection.
</Note>

## Summarization Prompt

When using **Warm (Summary)** mode, the AI automatically generates a conversation summary to brief the receiving agent. By default, the AI uses a built-in prompt to create this summary.

You can provide a **custom summarization prompt** to control exactly what information is included and how it's structured. This is useful when you want the briefing to focus on specific details, such as account numbers, issue categories, or caller sentiment, rather than a generic recap. Configure this in the transfer action settings or via the API.

***

## Call Screening

Call Screening lets the transfer recipient speak with the AI agent before being connected to the caller. Instead of an immediate connection, the recipient has a two-way conversation with the AI to evaluate the transfer.

During screening, the receiving agent can:

* Ask the AI about the caller's issue, sentiment, or any details from the conversation
* Decide whether to **accept** or **decline** the transfer

The caller hears hold music while screening takes place. If the recipient declines, the AI resumes the conversation with the caller and can attempt a fallback action, such as taking a message, trying a different transfer destination, or scheduling a callback.

<Note>
  Call Screening works with all warm transfer modes (Message and Summary). It is not available for blind (cold) transfers.
</Note>

***

## Preserve Original Caller ID

By default, warm transfers show the Synthflow agent's phone number as the caller ID to the transfer target. Enable **Preserve Original Caller ID** to display the **original caller's phone number** instead. See [How does Caller ID work?](#how-does-caller-id-work) for a full breakdown by transfer mode.

<Warning>
  Preserve Original Caller ID only works with phone numbers purchased through the **Twilio integration inside Synthflow** (i.e. numbers bought directly from the Synthflow dashboard). If your number was purchased outside of Synthflow, through your own Twilio account, a third-party provider, or imported via custom SIP, this feature is not supported and the transfer attempt will fail.
</Warning>

***

## **Advanced Settings**

In the **Advanced Settings** section, you can fine-tune the behavior of the call transfer action:

| Setting                     | Description                                                                                                                                              |
| --------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Extension**               | DTMF digits to dial after connecting (e.g., `123` or `123#`). Supports 0-9, \*, and #.                                                                   |
| **Timeout**                 | How long (in seconds) to wait for the agent to answer. Default: 30 seconds.                                                                              |
| **Human Detection Timeout** | How long to wait for a live human to be detected before treating the transfer as failed. Only applies to warm transfers. Min 30 seconds, max 30 minutes. |
| **Pre-message pause**       | Delay before the whisper message plays, giving the agent time to settle.                                                                                 |
| **Post-message pause**      | Delay after the whisper message, before connecting the caller.                                                                                           |

***

## Hold Music

Enable background music to keep callers engaged while waiting for the transfer to complete. This provides a professional experience and reassures callers that they haven't been disconnected.

To enable hold music:

1. Toggle **Background Music** in the transfer action settings
2. The default hold music will play while the recipient's phone rings

<Note>
  Hold music stops automatically when the receiving party answers and the call is connected.
</Note>

## Record the Call

By default, the recording stops once the call is transferred. If you want to continue recording the call, set the <code>stop\_recording\_after\_transfer</code> parameter to false.

```shell curl
curl --location --request PUT 'https://api.synthflow.ai/v2/actions/<action_id>' \ 
--header 'Authorization: Bearer <api_key> \ 
--header 'Content-Type: application/json' \ 
--data '{ "LIVE_TRANSFER": { "stop_recording_after_transfer": false 
} 
}'
```

***

## Transfer Failure Handling

What happens when a transfer fails depends on the transfer mode:

| Failure Scenario    | Blind Transfer | Warm Transfer                     |
| ------------------- | -------------- | --------------------------------- |
| No answer (timeout) | Call ends      | Agent can retry or continue       |
| Busy signal         | Call ends      | Agent can retry or continue       |
| Invalid number      | Call ends      | Agent notified, can try alternate |
| Network error       | Call ends      | Agent can retry or continue       |

<Warning>
  **Blind transfers cannot recover from failures.** If reliability is critical, use warm transfers to allow graceful fallback behavior.
</Warning>

For warm transfers, you can configure fallback behavior:

* Have the agent apologize and offer to take a message
* Try an alternate transfer destination
* Schedule a callback for later

***

## **FAQs**

### How does Caller ID work?

| Transfer Mode          | Caller ID Shown          | SIP Method Used |
| ---------------------- | ------------------------ | --------------- |
| Blind (Cold)           | Original caller's number | SIP REFER       |
| Warm (Message/Summary) | Synthflow agent's number | SIP DIAL        |

**Why the difference?** Blind transfers use SIP REFER, which hands off the call entirely to your telephony provider, preserving the original caller ID. Warm transfers create a new outbound call leg (SIP DIAL) from Synthflow, so the agent's number is shown. To override this for warm transfers, see [Preserve Original Caller ID](#preserve-original-caller-id).

<Note>
  If you're transferring to a PBX or using a custom SIP trunk, ensure your system accepts SIP REFER requests for blind transfers to work correctly.
</Note>

### What happens if a blind transfer fails?

The call between the AI agent and the customer ends immediately. There is no opportunity to retry or recover. If you need failure recovery, use a warm transfer instead.

### Can I pass caller information to the receiving agent?

Yes, with warm transfers:

* **Whisper message:** Speak a custom message only the agent hears
* **Context summary:** AI generates a summary of the conversation
* **Custom SIP headers:** Pass structured data to your PBX (for SIP transfers)