Continuing on with the creation of my HR Copilot series. In this blogpost I will show you how to set up generative AI capabilities on your own data in Copilot Studio. I want our chatbot to answer questions found in our HR documentation. This will be the starting point of our Copilot and instantly make it useful!
To reiterate we are building a HR Copilot to use inside Microsoft Teams. We want to solve the following problem statements with our Copilot:
- As an employee I want an easy process to call in sick.
- As a (project) manager I want to be notified when a team member is sick.
- As an employee I want to know my holiday balance and request time off.
- As a manager I want to be able to approve time off requests of my team member.
- As a HR agent I want help in answering HR related questions.
- As a HR agent I want to be notified when questions can’t be answered by the Copilot.
Now in this blogpost we will tackle the fifth bullet of the list. We will add immediate intelligence to our Copilot by connecting it to our existing HR handbooks.
Start with data
If you want to build along (please do!) I assume you have already created a solution and the HR Copilot. If you want to have a couple more pointers on how to do so, please look in a previous blogpost of mine here, or for official documentation go here.
We are going to use generative AI to answer employees question. And what is most important for generative AI? Yes, you know it, correct Data! If you want to let your HR Copilot scan your HR documentation, make sure it’s the correct version. Also check if there are no conflicting documents which have different nuances.
Now the good thing is that you can point the HR Copilot to a specific SharePoint library. This can be the place where you only publish the documentation that you want your HR Copilot to be able to answer. You can create subsets of documents and leave out conflicting information or information that you rather tell in person to your employees instead of letting the Copilot answer that question.
Do make sure the documents you enter are not too unusually large, it will be harder and maybe impossible to scan through all the text. You can split the document in multiple files if you want though!
In my example I use a SharePoint library with a single HR example handbook inside it. If you want to build along in your own developer environment, you can find the handbook here.
Point to your website
During the creation of your Copilot you could have already connected it to your SharePoint Library. This is the most easy way to add incredible intelligence to your virtual agent. If you have skipped as I have you now also have this highlight inside your starting overview.
You might need to turn on the “Boost conversational coverage with generative answers” toggle in the Generative AI tab. Else you will not have the option to enter the SharePoint Library Url.
And now enter your website, save your settings and your chatbot will now be able to answer the questions in your handbook.
Voila you are already done! Or at least on the Copilot Studio side, to deploy the Copilot to Microsoft Teams we need to add manual authentication.
Add authentication before publishing
The best part of using a SharePoint Library as the source for you data? It will use the credentials of the user interacting with your Copilot to search your SharePoint Library. So if a user does not have rights to read a document, the Copilot also cannot read the document!
Unfortunately to set this up we will need to add an application registration in Microsoft Entra Id. I say unfortunately as this is a bit more technical then all the previous steps. However there is ample documentation available on how to set this up on the Microsoft Learn site here.
If you only have a couple smaller documents that can be made available for all users interacting with your Copilot. It is also an option to upload these specific documents in Copilot Studio. This way you can use the “Only for teams” authentication option. However, I personally like that I can give the HR department control and the responsibility of what documents can be scanned by the Copilot in the SharePoint Library.
I find it personally very impressive how fast we can create a useful Copilot! Hope you find this little walkthrough useful, I hope I put enough of my personal experience and pointers to it that it can be beneficiary to even the more experienced Copilot builders!