# AI & Human Agents For Non Interactive Replies

## Introduction

In this guide, we explain how **Picky Assist** intelligently handles **non-interactive replies** using **AI Agents** or **Human Agents (Teambox)** within a chatbot flow.

This ensures that conversations remain smooth, natural, and uninterrupted—even when users respond in unexpected ways.

### What Are Non-Interactive Replies?

A **non-interactive reply** occurs when a user is shown an interactive message and is expected to click a button or option, but instead responds by typing or sending a voice message.

#### Common Interactive Elements

* Buttons
* List Menus
* WhatsApp Flows
* WhatsApp Payments
* WhatsApp Cart

Normally, these elements require the user to **tap** to move to the next step. However, real users often prefer to **ask questions** or **clarify doubts** before proceeding.

Picky Assist is designed to handle this behavior seamlessly.

### How Picky Assist Handles Non-Interactive Replies

When a user sends a **text or voice message** instead of clicking an expected interactive element:

1. The system **detects the non-interactive reply**
2. The conversation is **contextually routed** to:
   * an **AI Agent**, or
   * a **Human Agent (via Teambox)**
3. The agent responds to the user’s query naturally
4. Once the doubt is resolved, the chatbot **automatically redirects** the user back to the correct interactive step

This approach ensures:

* No break in the chatbot flow
* No user confusion
* A more human-like experience

### Example 1: Payment Flow with User Questions

#### Scenario

A user is shopping online and reaches the **payment stage** of a chatbot flow.

The chatbot displays a **WhatsApp Payment button** to complete the transaction.

#### Expected Action

👉 User clicks the **Pay Now** button

#### Actual User Behavior

Instead of clicking, the user types:

> “If I make the payment now, when will the product be shipped?”

#### What Happens Next

* Picky Assist detects that the user did not click the payment button
* The query is routed to:
  * an **AI Agent**, or
  * a **Human Agent**, depending on your configuration
* The agent replies with shipping details and delivery timelines
* After the response, the system **automatically re-shows the payment block**
* The user can now confidently click the payment button and complete the order

✅ Result: Higher payment completion and better user trust

### Example 2: WhatsApp Flow with Clarification

#### Scenario

A user is filling a **WhatsApp Flow** (form) for a service enquiry.

#### Expected Action

👉 User selects options inside the flow

#### Actual User Behavior

The user sends a voice message:

> “Can you explain what this service includes?”

#### What Happens Next

* The conversation is routed to an AI or Human Agent
* The agent explains the service details
* Once clarified, the chatbot **resumes the same WhatsApp Flow**
* The user continues filling the form without restarting

✅ Result: Zero drop-offs and a friction-free experience

### Why This Matters

This approach ensures that:

* Conversations feel **natural and human**
* Users are not forced to follow rigid flows
* Doubts are cleared instantly
* Conversions (payments, submissions, bookings) increase
* Automation and human support work **together**, not separately

### Key Benefits

* Seamless switch between **automation, AI, and human agents**
* No loss of chatbot context
* Higher user satisfaction
* Better conversion rates
* Ideal for commerce, support, bookings, and lead qualification

## How to Configure AI Agents in a Flow

This guide explains how to configure **AI Agents** inside a chatbot flow in **Picky Assist**, enabling intelligent handling of non-interactive and unexpected user replies.

### Getting Started

To configure AI Agents in a flow:

1. Open the **Flow Builder**
2. Locate the **AI Mate icon** on the **left side of the screen** (as shown below)\ <br>

   <figure><img src="/files/xgRS4EiWhLkIsZ90JW7m" alt=""><figcaption></figcaption></figure>
3. Click the icon to start configuring AI settings for your flow

{% hint style="warning" %}
⚠️ **Important:**\
Make sure you have already configured your **ChatGPT API token** before proceeding. [*Read More*](/setting-up-guide/setting-up-ad-ons/setting-up-chatgpt-app.md)
{% endhint %}

#### AI Provider

#### Supported Providers

* **Currently Supported:** ChatGPT
* **Coming Soon:** Gemini and additional AI providers

<figure><img src="/files/HNNG53rwmZA7AbOqeD6o" alt=""><figcaption></figcaption></figure>

#### Model Selection

Choose an AI model that best fits your use case.

**Recommended Model:**

* `gpt-4o-mini`
  * Suitable for most use cases
  * Cost-effective
  * Fast and reliable for conversational flows

<figure><img src="/files/PhOTTPFuXcudNTyFYGM6" alt=""><figcaption></figcaption></figure>

#### Instructions (AI Training Content)

This section is where you define **how the AI should behave and respond**.\ <br>

<figure><img src="/files/c7R6f3BJi4gefmNmFh3j" alt=""><figcaption></figcaption></figure>

You should include:

* The **role** of the AI (e.g., sales assistant, support agent)
* **Behavior and tone** (polite, professional, friendly, etc.)
* **Knowledge base** relevant to this specific flow
* **Boundaries** (what the AI should or should not answer)
* **Language preference**, if replies must be in a specific language

{% hint style="info" %}
💡 Tip:\
Always clearly define **role, behavior, knowledge scope, and boundaries** to get accurate and consistent responses from the AI Agent.
{% endhint %}

### Conversation Settings

These settings help businesses control AI behavior, cost, and flow continuity.

#### 1. AI Invocation Limit per Block

**Choose how many times the AI can respond within a single block.**

<figure><img src="/files/s3FV24evYkGRl6OQN44b" alt=""><figcaption></figcaption></figure>

Example:

* If set to **20**, the AI Agent can generate only **20 replies** in that block.

This helps:

* Control token usage
* Manage AI billing effectively\ <br>

#### 2. Resume Conversation After X Minutes (Inactivity)

This setting defines when the system should **return the user to the chatbot flow** after AI interaction.

<figure><img src="/files/MBjrY6sWjC8UZesRfeUt" alt=""><figcaption></figcaption></figure>

**Example:**

* If set to **2 minutes**
* If there is no response from the user for 2 minutes
* The system automatically switches from the AI Agent back to the flow
* The user continues from where they left off

This prevents conversations from getting stuck with the AI Agent.

#### 3. Resume Conversation If User Replies With Keywords

This allows the system to return to the flow when users respond with simple confirmations.

<figure><img src="/files/LlxFP1LGCj3Cyz32oH3k" alt=""><figcaption></figcaption></figure>

**Example Keywords:**

ok,done, yes, fine

You can add **comma-separated keywords**.

**How it works:**

* AI answers the user’s question
* User replies with “ok”
* The system automatically switches back to the flow
* The next chatbot step continues

#### 4. Resume Conversation When a Reaction Is Received

When enabled:

* If the user reacts (👍 ❤️ etc.) to the AI Agent’s message
* The system hands the conversation back to the flow
* The flow resumes from where it stopped

<figure><img src="/files/NGAFZUXk2l063brHVvGf" alt=""><figcaption></figcaption></figure>

#### 5. Resume Conversation When an Emoji Is Received

Some users respond with emojis instead of text.

When enabled:

* Emoji replies (👍 😊 👌 etc.) are detected
* The system resumes the chatbot flow automatically

### Enable AI assistance for all steps

When **enabled**, AI Assistance is automatically activated for **all flow steps** that support non-interactive reply handling.

<figure><img src="/files/z88huXkTc8NRMnMMKNGT" alt=""><figcaption></figcaption></figure>

#### Benefits

* No need to enable AI manually for each step
* Faster setup for standard use cases

#### Step-Level Override (Advanced Control)

Even with Global AI Assistance enabled, you can **override settings per step**.

This is useful for:

* Advanced flows
* Sensitive steps (payments, confirmations)
* Routing only specific blocks to AI

## Enabling Non-Interactive Handling per Step

For each interactive step in the Flow Builder, you can define how **non-interactive replies** should be handled.

If Global AI Assistance is enabled, **step-level preferences override global settings**.

#### Available Options

**1. Switch to Teambox (Human Agent)**

* Routes the conversation to a human agent
* Uses settings defined under **Handle Non Interactive Responses** under **Flow Settings**
* Ideal for complex or sensitive queries

**2. AI Assistance**

* Routes the conversation to the configured AI Agent
* Best for FAQs, clarifications, and quick responses

<figure><img src="/files/UQx4Rf77MK15DnhDsKEW" alt=""><figcaption></figcaption></figure>

## Handling Non-Interactive & Unexpected Replies with Human Agents

Now let’s see how to route conversations to **Human Agents** using **Picky Assist Teambox**.

### Handle Non Interactive Responses

1. Go to **Chatbot Settings**
2. Open **Handle Non Interactive Responses**

By default:

* **Assign Chat** action is enabled

You can also add additional actions such as:

* Sending email notifications
* Adding tags

<figure><img src="/files/1IGpu08HOBxqcVqYUGRG" alt=""><figcaption></figcaption></figure>

#### Resume Conversation After X Minutes (Teambox Inactivity)

This setting monitors inactivity after a chat is handed over to a human agent.

**Example:**

* If set to **15 minutes**
* Chat is transferred to Teambox
* If neither the user nor agent replies within 15 minutes
* The system automatically resumes the chatbot flow
* The user continues from where they left off

<figure><img src="/files/p9cqcdD55MSh728WDfnb" alt=""><figcaption></figcaption></figure>

### Do not resend the message block when resuming the chatbot conversation

When this setting is **enabled**, the system will **not resend the last message block** that was shown to the user before the conversation was handed over to an AI Agent or Human Agent.

Instead, the conversation is **directly returned to the chatbot flow**, allowing it to continue with the **next step** from where the user left off.

<figure><img src="/files/WQuDIeXUrWgwvMJ4sSyd" alt=""><figcaption></figcaption></figure>

### Identifying Flow-Routed Conversations in Teambox

Agents can easily identify chats routed from a flow by:

* A **special icon**
* A **countdown timer**

<figure><img src="/files/G41Ml5LsIFgo6o1TS0Zg" alt=""><figcaption></figcaption></figure>

This helps agents:

* Understand urgency
* Respond within the configured time window

Every new reply from the agent or user **resets the timer** based on your settings.

### Manually Routing Back to the Flow

In some cases:

* The agent resolves the user’s doubt quickly
* The user should immediately continue the chatbot journey<br>

  <figure><img src="/files/UeyVwcwKyVMgc6weHt8F" alt=""><figcaption></figcaption></figure>

Agents can manually:

* Transfer the conversation back to the flow
* Allow the user to resume steps without waiting for the inactivity timeout


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://help.pickyassist.com/setting-up-guide/setting-up-flow-builder/ai-and-human-agents-for-non-interactive-replies.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.