Building Your Own Chatbot Using Microsoft's New AI Hub (Part 1)

The AI Clinic: Deep Dive #2 | Subscribe

Editor’s Note: This tutorial wound up being longer than I expected so I’ve split it up into two parts — for your sanity and mine.

Every Tuesday, we take a closer look at the AI tools and trends shaping the future of healthcare. As artificial intelligence becomes more embedded in clinical settings, staying ahead of the curve isn't optional — it's essential for those who want to fully understand how these tools will be used in our industry going forward.

This week, we’re diving into Microsoft’s new Copilot Studio, which is making waves with its focus on healthcare-specific use cases. AI tools like Copilot Studio hope to bring the idea of personalized "AI agents" from fiction into reality. These agents aspire to perform tasks from routine administration to advanced clinical support, automating the more menial parts of frontline healthcare work.

So how do you actually set up one of these agents? And what real-world value do they bring to healthcare professionals?

In today’s Deep Dive, we’ll walk through configuring an AI agent using Microsoft’s platform, and explore practical applications. Let’s see how AI-driven tools can reduce admin burdens and let clinicians focus on what matters most: patient care.

How to Build Your AI Agent

So. Let’s walk through the steps for actually building an AI chatbot for ourselves using Microsoft’s Copilot Studio.

For today's walkthrough, I'm focusing on creating a Virtual Greeter that can be deployed to a website to answer basic questions and capture leads. This is the sort of tool that has universal appeal across hospital and outpatient service lines and should be straightforward enough.

This Virtual Greeter will answer basic questions about our business and collect patient information to pass along to us via email.

Here’s how to do it step by step:

1. Access Copilot Studio

   • Sign in to your Microsoft account and navigate to Copilot Studio via your Office 365 dashboard (or setup an account once you're there).

   • Once you sign in you should see the Copilot Studio dashboard/homepage

     

2. Create a New Agent

   • Click on the “Create” button on the left panel

   • Then select the "New copilot" option that pops up

     

3. Configure Your Chatbot

   • Next you’ll be on a screen that asks you to guide the setup of the assistant

   • Use the chat window to provide instructions for how you want your chatbot to function

     

   • Feel free to use the prompts that I provided below

     • Basic instructions: “Our copilot will assist patients by providing a warm greeting when they visit the medical practice's website, answering frequently asked questions such as clinic hours, services provided, and insurance accepted. Additionally, the copilot will collect essential patient information through an easy-to-complete form, helping the clinic streamline new patient intake and reduce administrative work.”

     • Tone: “The copilot should maintain a warm, professional, and friendly tone. It should be welcoming and helpful without being overly formal or rigid. We want patients to feel at ease and comfortable when interacting with it, as if they were speaking to a caring receptionist. The responses should be clear, empathetic, and succinct to ensure patients feel heard and supported.”

     • Topics to be avoided: “The copilot should strictly assist with medical practice-related tasks. It should not provide medical advice beyond general information about our services, nor should it attempt to handle sensitive health conditions, diagnoses, or personal medical data. Additionally, it should avoid any unrelated tasks, such as performing calculations, giving academic assistance (like doing math homework), or providing financial/legal advice. We want it to stay focused purely on guiding patients regarding our practice and services.”

   • Your chatbot will need a "knowledge source" to use to reference for basic questions about hours, services, and locations.

     • For this example, I used the website of a physician I know, Dr. Vipin Kuriachan (an excellent Ophthalmologist if you live in the Irving, TX area).

   • When you're satisfied with the basic configuration of your chatbot, hit the "Create" button in blue at the top right

     

4. Turn on Generative AI

   • You should now see the dashboard specific to our new chatbot

     

   • Next go to the "Settings" button on the top right

     • Click on "Generative AI" and click the bubble next to "Generative (preview)".

       

5. Setup a Topic for the Chatbot to Respond to

   • Go to the "Topics" tab and turn off all of the pre-loaded topics. They're templates and you can explore them at your leisure later. But for today they're only going to confuse our setup.

     

   • At the top left click on "Add a topic" and click "Create from description with Copilot"

   • On the next screen, name our topic as "Generate lead" and say that our topic is to "Let a user leave their full name and phone number". Then hit "Create" at the bottom.

6. Configure Your Topic

   • Using our inputs, copilot sets up a workflow for the chatbot to follow

     • When prompted by the user, the chatbot will follow a workflow to ask the user to leave their name and phone number.

       

   • We're going to make a few edits to make the experience more friendly for patients.

   • For the questions about getting the user's full name and phone number, we're going to use a friendlier prompt.

     • Describe what the topic does: "This tool helps patients when they want to schedule an appointment by taking basic information to call them back"

     • First question: "We'll have someone from our team reach out to you about scheduling an appointment. First, can you tell me your full name?"

     • Second question: "Thanks. What's a good phone number for you to be reached at during business hours?"

   • When you're done, hit "Save" at the top right.

7. Build Your Topics

   • Build as many additional topics as you'd like your chatbot to cover with users.

   • For this test version, I built a topic called "Basic Services" that informs patients what medical services Dr. Kuriachan's office has.

     

   • I also turned on the "Conversation Start" topic and gave it a canned text to start every chat with

     • "Hello, I'm (Bot Name)​, a virtual assistant. Just so you are aware, I sometimes use AI to answer your questions. I can help you schedule appointments or answer basic questions about the practice."

8. Test your Chatbot

   • Once satisfied with the topics, test out your chatbot using the chat window on the right.

   • Here's an example of our lead generation in action

     

   • And you can even test the knowledge center in action. Here's some "off script" question about the practice address where copilot pulls the information from the keitx.com website.

     

That's it for part 1!

Next week we'll cover configuring the chatbot to keep an spreadsheet of patients that leave their information, emailing us whenever a new patient contacts us, and how to deploy the chatbot to our website.

-Patrick

Enjoy this? Please take the time to forward to a friend or subscribe.