Create a Better, More Powerful No-Code QnAMaker Bot

I’ve blogged a few times about Question and Answer bots on my blog and have presented on QnA bots at several conferences around the globe.

I appreciate the technology and believe it has the ability to make users’ lives easier and less frustrating. Now, instead of having to know which page the information they are looking for is on, or having to remember the exact keyword to search for on the page, the user can use natural language to ask questions and QnAMaker can determine the users’ intent and return the correct response.

And it’s really easy to set up.

There are a couple of issues with setting up a standard no-code bot with QnAMaker though. This process usually involves setting up a bot service in Azure which can lead to more costs and it’s not easy or clear how to monitor the bot to know if it’s working well for users.

To be clear, a QnA bot is not simply something you deploy and walk away from. They take care and feeding. What questions are users asking? Do you have coverage for those questions? Is the bot responding with appropriate responses? How can you even tell?

How can you tell if your bot is doing its job well? If you are a developer this is easy, all you have to do is download the source code for your bot from the Azure Bot Service and modify the code to make it do whatever you want. However, what if you aren’t a developer? How can you make sure this QnAMaker bot is working well?

It turns out, you can use the QnAMaker API endpoint to create a bot using Power Apps and Power Automate quite easily. What’s more, you can add additional functionality to the Power Automate Flow to track what a user is asking and what the response from QnA Maker was to audit the bot’s usage and accuracy.

How is all of this done? Glad you asked. The video below walks you through the entire process of:

  • Creating a QnAMaker KnowledgeBase
  • Creating a Power App where users can ask questions
  • Creating a Flow to receive the question from the Power App and then send that question to the QnAMaker endpoint to get a response
  • Log the question, response, and the score that’s returned from QnAMaker in a SharePoint list for auditing
  • Responding with an “I don’t know” response if the score returned form QnAMaker is less than 50%.

As an extra added bonus, in the video I also run into the “…cannot be evaluated because property ‘shared_sharepointonline’ doesn’t exist,” error and show you how to get around that.

So, as you can see, with very little effort you can not only create a QnAMaker bot with no-code that doesn’t involve the Azure Bot Service, but you can also add more power to it by logging user questions and customizing the response. Think of the possibilities you can do by further enhancing the Flow and your KnowledgeBase.

I’m a big believer in bots, if you aren’t using them yet, give it a shot. Maybe you’ll be surprised by the results.

References:

1 Comment

  1. Hi Mark, THANK YOU!!! Question, is there a way to format the output to the PowerApp? For long answers and most that include hyperlinks it’s all smooshed up despite using “\n” to try and space it. Thanks!

Comments are closed.