> ## Documentation Index
> Fetch the complete documentation index at: https://help.agentsy.pro/llms.txt
> Use this file to discover all available pages before exploring further.

# Human handoff

> Know when and how to transition an AI conversation to a real person without interrupting the visitor's experience.

# Human handoff

An Agentsy AI Experience can help visitors explore questions, understand options, and become more confident about their next step. Sometimes that next step is a conversation with a person.

**Human handoff** is the transition from AI-assisted exploration to direct contact with the expert or business behind the experience. The goal is not to end the conversation as quickly as possible. It is to make human contact available **when it becomes useful, necessary, or explicitly requested**.

> [!NOTE]
>
> An AI Experience supports human relationships; it does not replace professional judgment, a live sales conversation, or services that require direct human involvement.

## When to recommend a human conversation

A handoff may be appropriate when a visitor:

* **Asks to speak with someone.** They request a consultation, meeting, callback, or direct contact.
* **Wants to take action.** They are ready to discuss a project, request a proposal, or explore working together.
* **Needs information the AI cannot verify.** They ask about live availability, inventory, current pricing, or account-specific details without a connected source of truth.
* **Needs personalized professional advice.** Their situation calls for a licensed professional or a qualified human decision-maker.
* **Reaches a limitation.** The assistant cannot safely or reliably resolve the request through its available instructions and knowledge.

Not every question is a handoff signal. A visitor asking "What does your service include?" may still be exploring. A visitor asking "Can I speak with someone about hiring you?" is making a much clearer request.

## Match the handoff to the visitor's intent

| Visitor intent | Recommended approach |
| - | - |
| Learning about a topic | Answer the question and continue helping. |
| Comparing options | Explain relevant differences and clarify only what is needed. |
| Exploring a possible fit | Help the visitor understand whether the service may be relevant. |
| Asking for a meeting or proposal | Offer the available booking or contact path. |
| Requesting unavailable live information | State the limitation and direct the visitor to someone who can verify it. |
| Seeking regulated or highly personalized advice | Provide appropriate general information and recommend a qualified professional. |

> [!WARNING]
>
> Do not invent meeting availability, promise a callback, imply that a message has been sent, or claim that a human has received the conversation unless the relevant action has actually occurred.

## Design a natural handoff

A helpful handoff generally has four parts:

**1. Recognize the visitor's request**

Identify whether the visitor wants information, personalized assistance, or a concrete next step. Avoid assuming that every expression of interest means they are ready to book.

**2. Acknowledge the context**

Briefly reflect what the visitor is trying to accomplish, without restating unnecessary personal details.

**3. Explain why a person would help**

Give a specific reason: reviewing available options, discussing a proposal, confirming details, or providing professional advice.

**4. Point to an actual next step**

Refer to the booking or contact option available on the AI Experience website. If no verified option exists, avoid inventing one.

### Example: consulting services

**Visitor:** "I think this approach could work for our team. Can we discuss implementing it?"

**Assistant:** "Yes. It sounds like you're ready to explore how this could work in your organization. A conversation with the consultant would be the best next step to review your goals and scope. You can use the contact or scheduling option on this page, if available."

### Example: boat advisory

**Visitor:** "Can you show me the boats you currently have for sale?"

**Assistant:** "I can help you clarify the type of boat that fits your needs, but I don't have verified access to current inventory. A Tu Bote Confiable consultant can review available options with you. You can use the scheduling option above the conversation to speak with the team."

This is a handoff because the request depends on **live inventory**, not because the visitor asked a difficult educational question.

### Example: insurance education

**Visitor:** "Which policy should I buy based on my medical history?"

**Assistant:** "I can explain how different types of coverage work, but choosing a policy for your personal circumstances requires a conversation with a licensed insurance professional. Please avoid sharing medical or other sensitive personal information in this chat. You can use the booking option on this page to discuss your options privately."

For experiences involving health, insurance, legal, or financial topics, define appropriate boundaries for the specific service and jurisdiction. Do not use the AI chat as a substitute for a secure process for sensitive information.

## Configure handoff behavior in system instructions

Your assistant's system instructions should describe:

1. **The triggers:** Which requests require or justify human involvement.
2. **The limitations:** What the assistant cannot access, verify, decide, or promise.
3. **The destination:** The actual booking or contact method offered by the AI Experience.
4. **The wording:** How to explain the transition in a helpful, non-pushy way.
5. **The continuation rule:** Whether and how the assistant should continue answering general questions after recommending contact.

Here is a reusable starting point. Adapt it to your business and the options actually available on your site.

```text theme={null}
HUMAN HANDOFF

Continue helping visitors with educational and exploratory questions.
Do not recommend a meeting solely because someone asks a question
or expresses general curiosity.

Recommend speaking with a person when the visitor:
- explicitly requests human contact;
- wants to discuss purchasing, engaging, or implementing a service;
- requests information requiring live systems or human verification; or
- needs individualized professional judgment beyond the assistant's scope.

Explain briefly why human assistance is appropriate.
Direct the visitor only to a verified booking or contact option.
Never invent appointment availability or claim that contact has
been made unless a supported action confirms it.

After recommending a handoff, continue answering appropriate
questions if the visitor wants to keep exploring.
```

> [!TIP]
>
> If your AI Experience has a scheduling button above the chat, you can tell visitors where to find it. Verify that the button is visible and works on both desktop and mobile before including that direction in your instructions.

## Keep helping after the handoff

A recommendation to book a meeting should not automatically end the AI conversation.

For example, after receiving a booking suggestion, a visitor may ask:

* "What should I prepare for the meeting?"
* "Can you explain the difference between these options first?"
* "What does the consultation usually cover?"

The assistant should continue helping with questions within its scope. Avoid repeating the same booking suggestion in every response.

## Handle visitor information responsibly

Handoff conversations can reveal useful context, but they may also contain personal information.

* Ask only for details needed to help the visitor at the current stage.
* Do not request sensitive medical, financial, identity, or other confidential information through an unsuitable chat flow.
* Do not assume a registered visitor has consented to every form of sales outreach.
* Follow the privacy notices, access settings, and applicable requirements for your AI Experience.
* Do not claim that conversation context will automatically be sent to a human unless that workflow is actually configured and verified.

## Test your handoff before launch

Use the following scenarios to check the assistant's behavior.

| Test prompt | Expected behavior |
| - | - |
| "I'm just learning about your approach." | Provides information without pushing a meeting. |
| "Can I book a consultation?" | Points to the real booking or contact option. |
| "What appointments are available tomorrow?" | Does not invent availability without a connected scheduling source. |
| "Can you show me your current inventory?" | States the live-data limitation and recommends human verification. |
| "I need advice based on private medical details." | Avoids collecting sensitive information and recommends an appropriate professional channel. |
| "Before I book, can you explain the process?" | Continues answering the question instead of repeating the handoff. |

### Handoff readiness checklist

* [ ] Handoff triggers reflect the business's actual services and limitations.
* [ ] The assistant does not recommend a meeting too early.
* [ ] The assistant recognizes explicit requests to speak with a person.
* [ ] Booking and contact links are real, accessible, and functional.
* [ ] The assistant does not invent appointment times or claim a booking occurred.
* [ ] Sensitive information is handled according to the experience's privacy requirements.
* [ ] The assistant remains helpful after suggesting human contact.
* [ ] The flow works on desktop and mobile.

## Related guides

* [Understanding user intent](/conversations-and-leads/understanding-user-intent)
* [Lead identification](/conversations-and-leads/lead-identification)
* [Conversation insights](/conversations-and-leads/conversation-insights)
* [System instructions](/create-and-configure/system-instructions)
* [Calls to action](/branding-and-website/calls-to-action)
* [Testing your assistant](/create-and-configure/testing-your-assistant)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.