# FAQ bot training and automatic replies

Train FAQ bot responses with operator feedback and control when an intent can reply automatically or start a task.

*Release 3.3 · admin, operator, supervisor*

**Settings › Bots › FAQ Bots**

## What it is

**FAQ Bots** match visitor messages to trained intents. An intent can suggest an answer to an operator, reply automatically when approved, or start a **Task**. Operator feedback helps INSIDE decide which intents are ready for automatic replies.

## Who uses it and why

- Operators assess suggestions while handling conversations.
- Supervisors with the **Bot Trainer** role review performance and pause unsuitable automatic responses.
- Administrators with the **Bot Trainer** role maintain training phrases, responses and intent thresholds.

## Where to find it

Open **Settings › Bots › FAQ Bots**, then select **Train** on the bot. The training interface includes intent content and performance views. Bot registration changes are Powerfront-managed.

## How it works

1. Add example visitor messages under **User says** and choose the response. You can write **Text Response**, select a **Saved Response** or select a **Task**.
2. While an intent is **In Training**, operators assess its suggestions. **Go Ahead** approves the proposed response; copying lets the operator use it in their own answer. **Not Helpful**, **Incorrect** and ignored suggestions have different reporting effects.
3. INSIDE calculates performance from recent eligible operator feedback after the last reset. Approval contributes more positively than copying; **Not Helpful** contributes negatively.
4. Once the sample is complete, a score at or above **Auto Reply Threshold** can move a training intent to **Auto-Reply**. A score below **Assistance Threshold** can block it.
5. Pause an intent to stop its response, or use **Reset Performance Score** to begin a fresh assessment.

Visitors see sent responses and any task that starts. Suggestions awaiting operator action are for operators. A matched intent can also set conversational context or mute further bot suggestions.

## Configure it

Use the **Bot Trainer** role to edit the intent's phrases, response, language versions and thresholds. Ask Powerfront to change the bot-level defaults or registration.

The bot fallback sample is 50, with an automatic-reply threshold of 90 percent. The assistance fallback is 10 percent, but the new-bot editor starts at 60 percent. Review the saved values before enabling automated replies. Intent-specific values can override the bot values.

Use a sample quantity of zero only when you intend to approve the intent without waiting for operator training. Pause or reset an automatic intent when its answer needs renewed review.

### Fields

| Field | Where | Required | What it is for |
|---|---|---|---|
| Intent name | Settings › Bots › FAQ Bots | Yes | Names the visitor intention represented by the training phrases and response. |
| User says | Settings › Bots › FAQ Bots | No | Supplies examples of visitor messages that should match the intent. |
| Text Response | Settings › Bots › FAQ Bots | No | Supplies the response, with alternatives to select a Saved Response or Task. |
| Auto Reply Threshold | Settings › Bots › FAQ Bots | No | Sets the performance score needed to leave training for automatic replies once enough feedback is available. The bot fallback is 90 percent; an intent can override it. |
| Assistance Threshold | Settings › Bots › FAQ Bots | No | Sets the lower performance boundary below which a trained intent is blocked. The bot fallback is 10 percent, while the new-bot editor initially shows 60 percent. Check the saved bot and intent values. |
| Performance Score Sample Quantity | Settings › Bots › FAQ Bots | No | Sets the recent operator-feedback sample used for training. The bot fallback is 50; the editor accepts 0 to 100. Zero can approve an intent without waiting for a feedback sample. |
| Mute Bot | Settings › Bots › FAQ Bots | No | Mutes bot suggestions for the conversation when the intent triggers. New intents initially leave this off. |
| Don't Respond | Settings › Bots › FAQ Bots | No | Allows an intent to match without supplying a text response, for example to absorb an unwanted match. |

### Permissions

| Permission | Lets you |
|---|---|
| Bot Trainer | Open FAQ Bots and train, pause, resume and reset intents. |

## Limits and good to know

- Matching a phrase does not guarantee a response. Bot eligibility, language, schedule, conversation mute state and intent pause state also matter.
- The feedback calculation uses at most 100 recent eligible records. Dismissed suggestions and feedback before the last reset do not complete the sample. **Incorrect** is tracked separately and does not count towards the required approval, copy and not-helpful total.
- Automatic approval does not guarantee that later negative feedback will disable an intent. Review and pause unsuitable responses.
- A task-triggered intent lookup only uses eligible intents approved for automatic handling. An intent still in training may work as an operator suggestion but not as a task answer.
- A selected task is an action, so a match may change the visitor journey instead of sending ordinary text. Task responses require Virtual Assistant capability.
- Current performance and a selected date range answer different questions. Do not expect their counts or scores to match.

## Depends on and works with

### This feature uses

- **Requires** [Custom bots and FAQ bots](ENG-020.md): The FAQ bot registration and its eligibility determine whether intent matching participates.
- **Requires** [Messages and rich content](CHAT-002.md): Matches produce operator suggestions or visitor-facing messages in conversations.
- **Configured in** [Virtual Assistant task setup](ENG-009.md): Task definitions supply the steps run by task-based intent responses.
- **Configured in** [Saved responses](CHAT-015.md): A selected Saved Response supplies the wording used by an intent.
- **Works better with** [Virtual Assistant task steps and branches](ENG-011.md): Tasks can request eligible automatically approved intent responses.

### Used by

- **Configures** [Custom bots and FAQ bots](ENG-020.md): Intent training defines FAQ responses and their automatic-reply thresholds.
- **Enhances** [Console keyboard shortcuts](CON-026.md): Bot suggestion actions can be bound to keys.
- **Enhances** [Chat tasks and topics for visitors](VX-017.md): Intent responses can supply related topic suggestions.

## FAQ and troubleshooting

**Why is an intent still In Training?**

Check the required sample and eligible operator feedback. Copying, dismissing and reporting **Incorrect** do not have the same effect as **Go Ahead**.

**Why does an approved intent not answer?**

Check whether it is paused, the conversation is muted, and a response exists in the visitor's language. Ask Powerfront to review bot targeting and scheduling when needed.

**Why did a response start a task?**

The intent has a **Task** selected as its response. Review that selection and the task's starting step.

**How do I stop an unsuitable automatic answer?**

Pause the intent. Correct its content and use **Reset Performance Score** if it should return to training.

**Why can I train an intent but not edit the bot registration?**

The **Bot Trainer** role provides intent-training access. Powerfront manages the registration and its default thresholds.