Skip to main content

Notifications

Announcements

Welcome to the Power Platform Communities - Lets Get Started!
News and Announcements icon
Community site session details

Community site session details

Session Id :
Power Platform Community / Forums / Copilot Studio - Topic Creation & Management / Using data from datave...
Copilot Studio - Topic Creation & Management
Unanswered

Using data from dataverse in to answer questions

(1) ShareShare
ReportReport
Posted on by Arjan 41

Hi everyone,

I've been struggling quite a bit over the last few days trying to build an intelligent bot that can provide users with project information stored in Dataverse. What I’m aiming for is fairly straightforward in concept:

When a user asks for a project status, I want the bot to retrieve and display the following details from Dataverse (which is available in Knowledge):


  • Project Name

  • Project Manager

  • Project Phases / Percentage Completed

 

Here’s what I’ve tried so far:

  • Adding instructions in the general instructions section

  • Creating a topic with two tools/prompts: one to determine the account/project, and another to handle the response format

 

Despite this, I keep running into issues. Sometimes the bot works and provides an answer, but then the next day it loops or throws errors. I’ve tried several variations, but nothing seems to be stable or consistent.

I understand this might be a broad or complex issue, but I’m mainly curious:

Has anyone here successfully pulled structured data from Dataverse into a bot and presented it in a defined format?

If so, how did you go about it? Any tips, patterns, or working setups would be much appreciated.

Thanks in advance!

 
Regards,
Arjan
Categories:
General
  • Suggested answer
    Romain The Low-Code Bearded Bear Profile Picture
    Romain The Low-Code... 265 on at
    Using data from dataverse in to answer questions
    Hello, i m working on a case with dataverse at the moment.
     
    Important point before trying tips : 
     
    - RAG is not "query on a database" : dataverse current RAG will only use the 10 best element to calculate answer. So everytime you try something with more than 10 element, the result will be wrong.
    - RAG is designed with GPT 4o (actual model) , so go to pick an answer in the middle of the database is not what it is build for :( .
     
    With that in mind, what cool thing you must be able to achieve (since i achieve them this moring)
     
    - Ask about all project of a manager (since under 10 of them)
    - make an action to calculate complicate things and ask for sample : "Give me all project from manager "xxxxx" and then call the action "calculate something cool" for each line. : it work fine: i did it with  a double task : power automate flow and then a free action to send mail with the result
     
    What it will failed by design : everything about date, order : GPT4o don't know when we are, you have to tell hime with function Now() in a message.
     
    About general instruction 
    - they are great to use with an autonomous trigger or deep reasoning (very great, use it often)  But it's a just "a pre prompt" so don't esparate it will solve and thing probleme like an advanced robot.
     
    Recommandation : use local prompt in your topics to adjust result, but ATM be aware that a prompt instruction is not a "code" instruction, genAI are here to generate, not to think deeplly ATM : use deep reasoning for this.

    About Dataverse strange problem
    - in a similar case i had a column with "project skills" : in a project i had "copilot studio, power platform", another on "power platform, copilot studio" : i ask : give me two project about power platform : dataverse say "no" only one (the one with power platform first) : why : because it's rag, not a query of unstructured data

    Quick answer to solde the problem : 
     
    -temporary don't use dataverse directly but use the new hidden feature sharepoint in dataverse (just under the button) with an excel file of your data
     
    Aucune description alternative pour cette image
     
    It will create behind the scene a dataverse but the index will be azure ai search with unstructured skill + semantic search (for still 2 messages consumption and the dataverse storage you use for the little file)
     
    it's very impressive and better.
     
    Hope for the futur 
    1. dataverse search will have a lot improvement :) be patient
    2. Dataverse as MCP server is comming and it's awesome
    3. GPT 4.1 is incoming very fast :)
    4. AI foundry integration is great for last resort solution but cost more.
     
    Hope i help you :) Please click the answer button if it help, it s important

    PS : if you speak french or a not afraid to read subbtilte, in my profil i have a youtube channel with a tutorial on how to use the new feature above :)
     

Under review

Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.

Helpful resources

News and Announcements

Welcome to the Power Platform Communities - Lets Get Started!

Quick Links

Michael Gernaey – Community Spotlight

We are honored to recognize Michael Gernaey as our June 2025 Community…

Congratulations to the May Top 10 Community Leaders!

These are the community rock stars!

Announcing the Engage with the Community forum!

This forum is your space to connect, share, and grow!

Leaderboard >

Last month Overall leaderboard

Featured topics

Product updates

Microsoft Power Platform Community release plans