• Greet
  • Be friendly and concise, you want customers to express their experience. Make them feel they are in a safe place. Explain that feedback is collected anonymously.
• Listen & Understand
  • Process a customer feedback
    • Detect the satisfaction
    • Detect improvement suggestion
    • Classify it into a topic such as Agent Experience, Agent Friendliness, Service Speed, Accuracy, Data Privacy etc.
• Automate the process of filling out the survey 
  • In our example we send the data to a SharePoint list
• Exit
  • Once one single message is processed, send the exit signal to Nimbus to leave the conversation.
  • If a customer does not want to give feedback, exit

• The Copilot needs access to SharePoint list. 
• As it only processes one message, it can terminate the conversation before sending the data to SharePoint.

1. Head to Copilot Studio
2. Creating a new (blank) Agent and describe it
3. Disable Orchestration

1. Add a tool to Copilot. The tool will use a Power Automate Flow to call an AI model and process the customer's input.

Tip: you can also replace this step by using the prompt tool in Copilot. It's up to you.

The Power Automate flow has 4 steps. which are detailed in the following.

The high level steps are:

1. The trigger is called when the agent calls the flow using the customer's message.
2. A HTTP request triggers your AI model to analyse the customer's message.
3. The response JSON is parsed and…
4. …returned back to the Copilot agent.

Details on each step will be explained in the following.

1. The flow trigger needs have a parameter defined which is the customer's input, type string. 
2. Name it CustomerInput.
    

The HTTP element will request an AI model of choice - in this case ChatGPT. 

• Fill in Request URI and Body as shown below
• ☝Don't forget to add your api-key. 

Body:

{
 "messages": [
  {
   "role": "system",
   "content": "You are an AI assistant that extracts customer feedback form a message."
  },
  {
   "role": "user",
   "content": "Given the following message:\n\n@{triggerBody()?['text']}\n\nRespond with a valid JSON:\n{\n \"topic\": \"...\",\n \"sentiment\": \"...\",\n \"room_for_improvement\": \"...\"\n}"
  }
 ],
 "temperature": 0.2,
 "max_tokens": 200
}

Parse the output of the AI model:

body('HTTP')?['choices'][0]['message']['content']

Once the Power Automate Flow is ready, the configuration of the tool is simple via the details:

1. Add the flow as a new tool to Copilot
2. Give it a name
3. Leave everything else default in the configuration. 
   (See screenshot).

1. Create a Custom Topic: “Process JSON Input”.
2. Pass along the TaskId for anonymous feedbacks. 
3. Add the phone number and customer's display name or any other data for non-anonymous feedbacks.

JSON to process

{
"TaskId" : "$(CallId)",
"Phone": "+$(Caller.TelNumber)",
"Name": "$(Customer.DisplayName)"
}

1. Create a custom topic “Gather Feedback”.
   This topic is only redirected to by other topics.
2. Add a Question to ask for feedback. Save response (Var1) as the input message for our AI Tool. 
3. Store the outputs into global variables.

1. Create the custom topic “Fill out the form”.
   This topic is only redirect to by other topics. 
2. Send the exit message first, then create the list item in SharePoint using the values from the global variables.
3. End this topic with “End conversation”.

You will also need a topic to greet the Customer. 

1. Create a topic “Greeting” which handles a greeting 
2. Add a redirect to the “Gather feedback” topic.

If your customer does not have any feedback, you need to handle this as well. 

1. Add another topic “No Feedback Wanted” which handles the wish not to leave feedback.
2. Leave a goodbye message and end the conversation.