AI Agent Guidance in BoldDesk is a structured set of instructions that defines how an AI Agent responds to customers. It controls the AI Agent’s role, instruction, Out of Scope, PII Governance, and Content Safety interactions across all support channels. Creating effective instructions for your AI agent ensures consistent performance, decision-making, and communication across all interactions. This article outlines key steps and considerations to help you define the role, responsibilities, and escalation process for your AI agent

Prerequisites

To configure AI Agent Guidance, you must meet the following requirements:

• Role: Admin or a role with permission to manage AI Agents (An AI Agent must already exist, or you must have permission to create one)
• Access path: Admin → AI Agent

When to Use AI Agent Guidance

AI Agent Guidance is useful whenever you need consistent and controlled AI behavior, including:

• Tier‑0 self‑service: Handling FAQs such as password resets or basic how‑to requests while escalating exceptions
• After‑hours support: Providing standardised responses and collecting required details for follow‑up
• Narrow‑scope assistants: Restricting the AI Agent to specific topics (for instance, shipping and returns only)
• Compliance‑driven responses: Preventing sensitive data requests and enforcing approved language
• Structured troubleshooting: Enforcing step‑by‑step responses with limited clarifying questions
• Workflow alignment: Ensuring required fields (order ID, priority, environment) are captured before next steps

How BoldDesk Applies Guidance

When generating a response, BoldDesk uses Guidance to determine:

• Whether the AI should answer directly or request missing details
• Which topics or content the AI must avoid
• How responses are structured and phrased

How to Add or Update AI Agent Guidance

1. Go to AI.

2. Select AI Agent and open the agent you want to configure.

3. Locate the Guidance section.
   
   

   
   
   

4. Customize the guidance fields, including:

   AI Agent Guidance	Description
   Agent Role	This section contains a text box where you define the overall identity, purpose, and behavioural expectations of the AI agent.
   Agent Instructions	This section provides space to write additional rules, constraints, or operational instructions for the agent.
   Immediate Handover	This text area is for defining when the AI should instantly hand off a conversation to a human agent.
   Conditional Handover	This section lets you specify situations where the AI should transfer the conversation to a human based on conditions or thresholds.
   Out of Scope	You can enable to ensure that the AI stays within the defined limits and does not respond to queries outside the allowed scope.
   PII Governance	If enabled, it enforces rules to protect personal and sensitive information in customer queries.
   Content Safety	Enables content safety checks to filter harmful or inappropriate material according to your preferences. You may select options for: Verbal abuse/harassment, Hate speech, or Dangerous/illegal requests.

5. Click Save/Update.
6. Test the Guidance using the preview tool labelled as playground.

Defining AI Agent Guidance

To guarantee the efficient operation of your AI Agent, please consider the following guidelines.

1. Defining the AI Agent Role

Clearly specify the role of your AI agent to ensure it operates within the intended scope. Consider:

• Purpose: Define the primary tasks the agent is expected to handle (for instance, customer support, data analysis, etc.).
• Boundaries: List areas where the agent should not intervene or act.

2. Creating AI Agent Instructions

Provide detailed instructions and set behavioural guidelines for your AI agent. Include:

• Response Style: Specify tone and language (for example, formal, friendly, or technical).
• Behaviour: Define how the agent should act in different scenarios (for instance, apologising for errors or thanking customers for feedback).
• Decision-making Criteria: Outline rules for making decisions autonomously.

3. Immediate Handover Criteria

Define scenarios where the AI agent must escalate issues immediately to a human representative. This prevents delays in critical situations.
Examples include:

• Technical problems beyond the agent’s understanding.
• Emotional or sensitive customer interactions.
• If no relevant documents are retrieved, respond exactly with: ‘I can’t find relevant information. Please add the relevant documents to the library so I can answer.’

4. Conditional Handover Guidelines

Set rules for conditional handovers when specific requirements or thresholds are met.
Examples:

• Complex or unsupported queries

  • If the issue involves multiple steps or requires human judgement
  • If documents exist but don’t answer the question, use the same fallback message.

• Customer Preference
  When a customer explicitly requests human assistance.

• Capability limitations
  If the request requires any capability beyond the knowledge base, do NOT attempt to answer.

Fixed Language Responses With AI Agent

BoldDesk AI Agent now allows you to set a preferred response language to ensure more consistent and accurate replies. When enabled, the AI Agent prioritises this configured language even when customer messages contain mixed content such as product names, codes, or technical terms. This is customised in the AI Center > AI Agent guidance under Agent instructions. You can add specific instructions to the AI Agent to prioritise the preferred language. This will help improve the AI Agent’s language detection and ensure that the responses are aligned with the end user’s language.

You can further improve accuracy by adding custom instructions that guide how the AI Agent interprets messages with mixed elements, helping it focus on the user’s intended language and deliver responses that better match customer expectations.

Recommended Guidance Elements

• Allowed and disallowed topics
• Train your AI agent to identify conditions for handover.
• Fallback behaviour when information is missing
• Escalation and handover rules
• Log handover scenarios to improve AI performance and refine the process.
• Required compliance language (if applicable)
• Continuous improvement support

Permission Required to Manage AI Agents

To manage AI Agents, users must have Manage AI Agents permission.

Go to Admin > AI > Roles and Permissions > Manage AI Agents

Troubleshooting AI Agent Guidance

The AI Agent ignores part of the guidance

• Remove contradictory rules.
• Rewrite guidance as brief, direct statements.
• Organise into sections: Scope, Escalation, Formatting.
• Clearly state restrictions (for instance, “Do not answer billing questions”).

The AI Agent asks too many questions

Add an instruction such as: “Ask only the minimum required clarifying questions; otherwise provide the best next step.”

The AI Agent answers out-of-scope topics

• Include explicit out-of-scope response and escalation rules.
• Provide examples of out-of-scope requests and handling.

Responses are too long or too short

• Specify a target length (for example, “Keep responses under 120 words unless more detail is requested.”)
• Require a structure (for example, “Summary → Steps → Next action”).

Frequently Asked Questions

1. When does a guidance change take effect?
   After selecting Save/Update, AI Agent Guidance applies only to responses generated after the save.

2. Does AI Agent Guidance override BoldDesk permissions or access controls?
   No. AI Agent Guidance does not override user permissions or system-level access controls.

3. Do I have to configure Out of Scope, PII Governance, or Content Safety?
   No. These controls are optional, but if any of them are enabled, a custom immediate handover or decline message is required.

4. What should I do if the AI Agent cannot find an answer in connected knowledge sources?
   Define escalation behaviour in AI Agent Guidance (for example, handover to a human agent). The AI Agent cannot invent an answer when no answer exists in connected sources.

Related Articles

1. How to Create and Configure BoldDesk AI Agent
2. How to Customise and Test an AI Agent in BoldDesk