Creating an AI Agent

Last updated on 


On this page you’ll find detailed information on how to train, deploy, and manage your AI Agents on Cue, as well as information on billing and data privacy.

How to create an AI Agent

To create an AI Agent for your Cue, do the following:

  1. Sign in to app.cuedesk.com
  2. Click on the AI Agents icon in the left nav
  3. Click on 'Create AI Agent'
  4. Give your AI Agent a name and prompt
  • Want to learn how to write a great prompt? Here is a handy guide to help you!

5. Provide training knowledge (websites & docs)

  • AI Agents can be trained on a variety of documents such PDF, DOCX, PPTX, and JSON, as well as your website.
  • Training takes approximately 1 - 3 seconds per single page document.

Testing an AI Agent

Once an AI Agent has been trained on the content you've uploaded you can test it using the live preview on the right of it's settings page (this preview only does one shot answers)

💡
AI Agents can only answer questions based on static training information. They cannot yet perform actions such as update a customer's details, or check the status of an order.

Deploying an AI Agent

If you would like to activate your AI Agent, please contact support@cuedesk.com or your account manager and they will arrange for it to be activated for your workspace

Once it has been activated you will be able to add a "Transfer to AI Agent" step to your flow, when a contact reaches this step they will begin conversing with the AI Agent

How AI Agents handle queries

The AI Agent will attempt to answer the customer's query by searching through the documents it has been trained on and composing an answer based on the most relevant snippets of information returned in the search results.

The AI Agent will at some point transfer the conversation back to the flow and will provide a "reason" for transferring the conversation back, you can then take the user down different branches of a user journey depending on what reason the AI Agent has provided.

The possible reasons for the AI Agent transferring back to flows are:

  • confirmed_resolved - Customer has confirmed that the AI Agent has successfully resolved their query.
  • assumed_resolved - Customer did not respond after receiving an answer from the AI Agent. After 24 hours the AI Agent reviewed the conversation and assumed it was resolved successfully.
  • confirmed_unresolved - Customer has confirmed that the AI Agent has not resolved their query (the AI Agent will make up to three attempts to ask clarifying questions or provide an answer before transferring)
  • assumed_unresolved - Customer did not respond after receiving an answer from the AI Agent. After 24 hours the AI Agent reviewed the conversation and assumed it was not resolved.
  • human_requested - Customer asked to speak to a human
  • error_unresolved - The AI Agent could not respond to the customer. Usually this is because Open AI's API is having problems.

Agent Handoff

Our AI Agent is designed to handle customer queries efficiently. However, there may be times when a customer requests to speak with a human. You can enable agent handoff through our live chat/inbox functionality by configuring your setup correctly.

You have two options for handling these requests:

  1. Provide Contact Information – The AI Agent responds with a support email, phone number, or other contact details.
  2. Live Chat Handoff – The AI Agent transfers the customer to a human agent in the Cue live chat/inbox.

Option 1: Providing Contact Information

To configure the AI Agent to provide a support email or phone number, include an instruction in your prompt, such as:

👉 "If a customer requests a human, please respond with 'For further assistance, please email support@cuedesk.com'."

You can customise this response based on your preferred contact details.

Option 2: Live Chat Handoff

To enable AI-to-human handoff within the Cue live chat/inbox:

  1. Configure your flow to detect when a customer requests a human (this request is given by the customer who is 'speaking' to AI).
  2. Set a flow step that transfers the conversation from the AI assistant to a live agent.  This is done by using:
    2.1. If condition
    2.2. First Value: AIAgentName.Transfer
    2.3. Second Value: human_requested
    2.4. Add a flow transferring the customer to live chat
  3. Ensure your team is available to respond within the inbox.

Please see below an example of how to set up your flow to accommodate this:

How much do AI Agents cost?

You are only billed for confirmed and assumed resolutions.
Please refer to www.cuedesk.com/pricing for pricing information.

Data privacy

When a conversation is connected with an AI Agent, the AI Agent can see the entire message history of a conversation. This may include any personal data that a customer has provided in a conversation. We use Open AI models for our AI Agents and so the full message history for a conversation is shared with Open AI. Open AI does not train their models on conversational data shared with them.

How to Calculate Your Cost Savings with AI Agents

Understanding your cost per resolution can help you measure the efficiency of your customer support operations and compare it with AI Agents. Follow the steps below to calculate your savings.

Step 1: Gather the Required Data

You’ll need the following information:

✔ Number of agents handling customer queries
✔ Their salaries (monthly cost per agent)
✔ Number of tickets handled per month

Step 2: Calculate Your Cost Per Resolution

Use this formula:

Example Calculation

If you have 5 agents, each earning £2500 per month, and they handle 500 tickets per month, your cost per resolution is:

(5 × 2500) ÷ 500 = £25 per resolution

Step 3: Compare with Cue’s Cost Per Resolution

SA Pricing:

  • 100-message bundle: R9.99 per resolution
  • 1000-message bundle: R8.99 per resolution

UK Pricing:

  • 100-message bundle: £0.99 per resolution
  • 1000-message bundle: £0.89 per resolution