Instruct your AI Agent

This article explains how to write effective instructions for your AI Agent using the Behaviour and Scenarios tabs.

Clear instructions ensure your AI Agent responds correctly, takes the right actions, and escalates when needed.

The three layers of instruction

Your AI Agent works with three layers of instructions. Each layer has a specific role:

Layer 1 — Behaviour settings

Defines who your AI Agent is. This applies to every conversation.

  • Job description
  • Tone of voice
  • Company context

Layer 2 — Handover criteria & Additional instructions

These are your default rules. They apply when no Scenario is active.

Think of this as your fallback layer.

Layer 3 — Scenarios

These are process-specific instructions for handling particular situations.

Important: Scenarios always take priority over all other instructions.

If no Scenario applies, Layer 2 is used instead.

Behaviour: define how your AI Agent operates

The Behaviour tab controls how your AI Agent communicates, makes decisions, and handles conversations by default.

Job description

Define the role and responsibilities of your AI Agent.

Best practices

  • Be specific about what the AI Agent should handle
  • Clearly describe key tasks

Example

"Assist customers with order questions, returns, and product information."

Company context

Provide background information about your business.

Include:

  • What your company does
  • Who your customers are
  • Your products or services

This helps the AI Agent give more relevant responses.

Tone of voice

Define how your AI Agent communicates.

Best practices

  • Be explicit (e.g. friendly, professional, concise)
  • Include what to avoid

Example

"Friendly and helpful, but professional. Keep answers concise. Avoid slang or overly casual language."

Handover criteria

Handover criteria define when the AI Agent should escalate a conversation to a human.

By default, the AI Agent escalates when:

  • It cannot answer a question
  • The customer asks for a human

You can add additional rules that apply across all conversations.

Examples

  • E-commerce: Escalate on payment disputes or chargebacks
  • Hospitality (THL): Escalate on safety issues or emergencies

Important: If a Scenario defines escalation rules, the Scenario takes priority.

Collect customer details before handover

Define what information the AI Agent should collect before handing over.

The AI Agent will only ask for missing information. It won’t ask customers to repeat themselves.

Examples

  • E-commerce: order number, email, issue description
  • Hospitality: booking reference, guest name, request details

This ensures your team has full context immediately.

Additional instructions

Additional instructions are always-on rules that apply to every conversation.

Use them for general rules, not step-by-step processes.

Examples

  • Always address the customer by their first name
  • Never promise a refund without confirming the order number
  • Do not discuss competitor pricing
  • Always ask if there’s anything else you can help with

Important: When a Scenario is active, it overrides Additional instructions.

Scenarios: handle specific situations step by step

Scenarios define how your AI Agent handles a specific process from start to finish.

Use a Scenario when:

  • There are clear steps to follow
  • Specific information needs to be collected
  • Actions need to be taken (e.g. assign, label, API call)
  • There is a defined escalation path

Tip:
Each Scenario should cover one complete process. Avoid combining multiple use cases.

Common Scenario examples

E-commerce

  • Return or refund request
  • Damaged or wrong item
  • Order status enquiry
  • Discount code issues
  • Payment disputes

Hospitality (THL)

  • Booking modifications
  • Cancellations
  • Guest complaints
  • Late check-in / checkout
  • Amenity bookings

How to write a good Scenario

A strong Scenario includes:

  1. Trigger — when should the Scenario activate?
  2. Steps — what should happen, and in what order?
  3. Required information — what should be collected?
  4. Actions — what should the AI Agent do?

What the AI Agent can do in a Scenario

Within a Scenario, your AI Agent can take actions in Trengo:

Assign to a team

"Assign the conversation to @Returns Team."

Assign to a user

"Assign to @Sarah."

Apply labels

"Apply the label @damaged-item."
"Apply the labels @refund-requested and @high-priority."

Add internal comments

"Add an internal comment with the order number, issue, and confirmation of photo received."

Use integrations or custom actions

"Use the Shopify Storefront to look up the product and include the price."
"Use the order status action to retrieve the order details."

The AI Agent will only use integrations if instructed or if clearly defined in the action setup.

Using @mentions in Scenarios

Use the @ button to insert:

  • Team members (e.g. @Sarah)
  • Teams (e.g. @Returns Team)
  • Labels (e.g. @damaged-item)

This avoids errors and ensures consistency.

Worked example — damaged item (e-commerce)

When a customer reports a damaged item:

  1. Apologize and acknowledge the issue
  2. Ask for a photo and order number
  3. Confirm the damage after reviewing the image
  4. Add an internal comment with details
  5. Apply the label @damaged-item
  6. Assign to @Fulfillment Team

Worked example — cancellation request (hospitality)

When a guest requests a cancellation:

  1. Ask for booking reference and reason
  2. Explain the cancellation policy
  3. Add an internal comment with details
  4. Apply the label @cancellation-request
  5. If within 48 hours of check-in, assign to @Sarah

Limits to keep in mind

  • Maximum 10 Scenarios per AI Agent
  • Maximum 5,000 characters per Scenario
  • Scenarios do not connect or chain together
  • The AI Agent only sees the current conversation
  • It does not know which Journey it belongs to

When to use what

Use the right instruction type for the right situation:

  • Additional instructions: rules that apply to every conversation
  • Handover criteria: situations that should always escalate
  • Scenarios: structured processes with steps and actions

Summary

To instruct your AI Agent effectively:

  • Use the Behaviour tab to define role, tone, and global rules
  • Use Scenarios for structured workflows and actions
  • Use Handover criteria and Additional instructions as fallback rules

Key takeaway: Scenarios always take priority. The clearer your instructions, the better your AI Agent will perform.

‍