Empowering Business Users With Databricks — Integrating AI/BI Genie With Microsoft Teams
Summary
- Seamless collaboration: Rooms To Go integrated AI/BI Genie into Microsoft Teams, enabling warehouse employees and sales teams to ask natural language questions and get data insights without leaving workflows.
- Mobile accessibility: Business users can explore data and run analyses directly from their phones via Teams, making real-time insights available to frontline staff and remote teams.
- Practical scaling: The integration demonstrates how organizations can democratize data access by connecting AI/BI to communication platforms teams already use daily for collaboration.
Empowering Business Users With Databricks — Integrating AI/BI Genie With Microsoft Teams

Talk by Nathan Sundararajan, Manager Data Engineering & Integration, Rooms To Go ; Ryan Bates, Solutions Architect, Databricks
In this video, we'll explore how Rooms To Go enhances organizational collaboration by integrating AI/BI Genie with Microsoft Teams. Genie enables warehouse employees and members of the sales team to interact with data using natural language, simplifying data exploration and analysis. By connecting Genie to Microsoft Teams, we bring real-time data insights directly to a user’s phone. We'll provide a comprehensive overview on setting up this integration as well as a demo of how the team uses it daily. Attendees will gain practical knowledge to implement this integration, empowering their teams to access and interact with data seamlessly within Microsoft Teams.
Intelligent analytics for everyone: https://www.databricks.com/product/business-intelligence
BI meets AI: https://www.databricks.com/resources/ebook/business-intelligence-meets-ai
Explore the only unified platform for agent systems: https://www.databricks.com/product/artificial-intelligence
See all the product announcements from Data + AI Summit: https://www.databricks.com/events/dataaisummit-2025-announcements
Chapters
0:00Introduction to Data1:07The Mouse and Cookie Tale1:57The Business Data Story4:00Democratizing Data with AI5:51Rooms To Go Use Case9:53Implementation Challenges12:53Technical Architecture15:04Genie Bot Configuration16:11Live Demo22:16Resources and Deployment23:28Future Roadmap and Security26:26Best Practices and QA
FAQs
How does Rooms To Go use AI/BI Genie in Microsoft Teams?
Teams members ask natural language questions about data in Microsoft Teams, and AI/BI Genie responds with insights, analytics, and visualizations—all without leaving the Teams interface.
Who are the primary users at Rooms To Go?
Warehouse employees and sales team members use the integration to access data on their phones, enabling frontline staff to make real-time decisions without IT involvement.
How can I give business users self-service access to data without relying on analysts for every question?
Business users can gain self-service access to data without analyst involvement by leveraging natural language AI tools like Databricks AI/BI Genie integrated into familiar platforms like Microsoft Teams. This video demonstrates how business users can ask data questions directly through their chat interface, receiving instant insights without writing SQL or requesting analyst support. The system handles data discovery, query generation, and result formatting automatically while maintaining governance and quality standards.
How can I give business users self-service analytics without sacrificing governance?
Giving business users self-service analytics without sacrificing governance requires a platform that combines flexible natural language access with centralized policy enforcement and audit trails. This video shows how AI/BI Genie integrated with Unity Catalog ensures that business users can only access data they are authorized to see, while all queries are logged for compliance and security. This balance enables organizations to democratize data access while maintaining enterprise-grade governance and regulatory compliance.
Which tools keep follow-up questions tied to the right business context?
Tools that keep follow-up questions tied to the right business context maintain conversation history and data context throughout extended interactions, ensuring coherent and relevant responses. This video demonstrates how AI/BI Genie in Microsoft Teams preserves context across multiple questions, allowing users to refine their inquiries and explore data naturally. By maintaining full conversation context and semantic understanding, the tool delivers insights that build logically on previous questions.
Full transcript
0:00 Introduction to Data
Okay. So, rule number one of presenting is do not present after lunch. And rule number two is do not present before happy hour. So, round of applause for all of you making it today. Uh. I'm a solution architect at Data Bricks. I work on our retail, CPG, and travel and hospitality team. Um, and just before we get started, I figured I would just talk about me a little bit since you probably aren't here to learn about empowering business users with data bricks.
You're probably here to learn about me. I hate reading. I hate reading so much. For me, was always a punishment in terms of schooling and uh my degree is in biomedical engineering and everything I read was boring and tedious. So, I really prefer picture books. And has anybody read the book If You Give a Mouse a Cookie? Show of hands. Okay, good. So, so some Yes, it's it is a children's book. That is correct. That's uh more my speed.
Um and so I wanted to take you all through that book today. So, if you give a mouse a cookie, so if you give a mouse a cookie, he will probably uh want a glass of milk. And when you give the mouse that glass of milk, he'll probably ask you for a straw. And then when he's finished, he'll ask you for a napkin. And then he'll look in the mirror to make sure he doesn't have a milk mustache. And this story goes on and on and on.
1:07 The Mouse and Cookie Tale
And he has a milk mustache. And he goes outside and he realizes the living room is dirty. So he has to get the broom. And when he has the broom, he realizes the dustpan is full. And then he has to take out the trash. And then he's tired. And the mouse has to take a nap. And at the end of his nap, the mouse wakes up and he realizes that he's thirsty. And he wants a glass of milk. And then what happens? This entire story starts all over again.
What does that have to do with what I'm talking about today? I want to tell you a separate story. If you give a business user a dashboard they like. Has anyone heard this story before? And with the power of generative AI and my charming personality, I'm going to tell you a new story today. So, if you give a business user a dashboard they like, they're probably going to want to have a filter on it. and you and your team are going to go build out the dimensions and understand what they're looking for and run it through all the cycles and provide this dashboard to them.
1:57 The Business Data Story
And when you give them that filter, they're going to realize that there is a custom metric missing that requires very specific custom logic and you and your team are going to go back and you're going to build that metric and you're going to QA it and make sure it's right and you're going to give it to them. And when that calculation shows up, they're going to get so excited and wonder if they can have that dashboard in real time.
And so now your team's going to go back and you're going to build a streaming data pipeline that takes all the data coming in from all the different sources and running this ETL as fast as you possibly can and it's going to be ready to go and it's going to be amazing. And when you give them that real- time dashboard, they're going to want to have alerts pushed to their mobile device so they can understand what's happening as soon as it happens.
So, your team's going to go back and understand, should I give them the alerts via email or how do we get it on their mobile device and who's going to get the alert and when? And now it's all ready to go. And when the alerts go off, they're going to say, "Man, I wish I could ask a follow-up question." And they're going to say, "It would be so great if we have a chatbot." And you and your team are going to do all this diligence and look through all these vendors and figure out what's the best way to harness this chatbot functionality that they came up with. and you are going to provide to them this amazing stunning piece of technology where you integrate data bricks and Microsoft teams into a chatbot that allows them to ask as many questions as they like in real time and if you give this business user this worldclass superpower they are going to ask you for five more dashboards and we're going to start this entire process over and over and over again.
Data Bricks, our mission is to democratize data and AI for everyone. And if you think about in your personal life, if there's any question that you possibly have, you pull out your phone and you ask your question. And that idea could not be more opposite in our business worlds. Any question that you have is siloed. And these systems don't communicate with one another. And this idea of what if you can now empower a business user to get these answers to their follow-up questions without waiting for an analyst, which is a really exciting thing, but it's something that we're so used to in our personal lives and we take for granted.
4:00 Democratizing Data with AI
Enter AIB Genie, which is essentially the idea of talking to your data and providing this self-service analytics that business users don't have. And even more importantly, how do I provide a very easy to use chat-like interface that's tailored to my business and governed and secure and ready to go that I can roll out to thousands of users with a push of a button. When you build out Genie Spaces, the idea is to tailor it to your data and your business.
It's designed around all of the dimensions and the tagging and the descriptions in your data set. It's tied to that user and what things they have access to. It's tied to previous questions that other business users may have come up with. And now you can iterate based on their feedback to say this is working and this isn't working to continue to improve this process. From a fundamental standpoint, how do we design very specific genie spaces for what a user wants to do?
And then how can I take that data and meet a business user where they're at? specifically in this example, Microsoft Teams, but it could be anything that you like. So, um is a data engineering manager from Rooms to Go, and that's exactly what and his team did. Thank you. And hey, uh everyone, thanks for coming to this meeting like said just before the happy hour. So, um I work for a company called Rooms to Go. It's predominantly in the east coast and we sell furniture and we have close to 160 stores and then we also have online presence.
5:51 Rooms To Go Use Case
So the first thing like made a valid point like you know we have PowerBI reports and it's extensively used but after the chatbot everyone started asking can I have like a bot kind of interface I want to ask questions. So now it became like you know the hot topic even though we have a cool dashboards and then the PowerBI they kept asking like I want more interactive kind of like a bot where it can answer questions and then they also want more like in real time.
So we thought like when Genie when data bricks came out with Genie, we like that interface and late last August or September they introduced the Genie API. We were one of the pioneers like you know we thought like we can use this and leverage that and then start using. So at the time was our solution architect and then you know I called him and then said I like the Genie API but I don't want to give my users access to the datab bricks interface.
I want them to be called from teams or slack. So he worked with some of the other architects and actually he wrote a blog and then which explains how this interface works and we tried it and then it was good. First technically let me tell the story. I wanted to give our executives the sales numbers because as a furniture store our sales they want to see the sales like they wanted to see it every 15 minutes. They want to see did I sell a sofa because it's a furniture store and you know technically you know if we sell a sofa every 15 minutes we will be in a different shape but so but they keep asking for like you know I want it and even though we have we gave them the PowerBI and it's in their phone again it comes to oh I have to click this button that button that filter so I want to ask questions.
So we created that you know the our first version of is the sales genie which used our sales numbers but the problem is with the powerbi we have role level security and all those implementation so like you know um not a each store cannot see the other stores sales numbers so with this integration at that time we were not able to do it but our team especially my team wants to show something that we can deliver. So we thought that it could be a good candidate for warehouse.
So we did this bot which is like you know answer in real time and then like you know familiar interface teams teams is everybody and even the warehouse guys has the teams in their phone. So we try to use that and then the important question is they want to have contextbased answers. In PowerBI you will have reports, you will have graphs everything but they want to have like talking to a person. They like that kind of interface.
So that's their main request. Two years back that PowerBI report was so good they said it meets our requirements and everything. Now they said like we want chatbot. So so we created this uh data which is our streaming we used a streaming live data which is using DTS and then we ingest the data in real time. So we track off all the inventories and then we will actually use that inter tables for this interface. So users want to ask questions and get answers in real time and then model this model will get learned based on the questions.
9:53 Implementation Challenges
When we first did it took us like three weeks or four weeks. We let the users the guys who are going to use come into the datab bricks interface ask questions then we looked at each question they are asking and then we looked at it and then the response whatever the genie gave then some of the response was like it was not right. So we actually created some SQL queries to train the model. Then we gave the instructions how to because some of the business terminologies like our guys use the warehouse number.
So we need to tell that warehouse number you know it has to interpret that how to interpret that like those kind of instructions. Then we built this instructions and then we used the API calls and then we used a Microsoft Azure bot. So when we deployed this first thing they said that even though we refresh every 15 minutes the current PowerBI reports still they are not happy. So they said oh I went all the way to that end of the warehouse and then I found out my inventory is not there because between that refresh time that inventory is gone.
So like you know 99 like 99 out of 100 times it might work only one time that inventory was not there they made it like you know they will put it in the teams and say like you know this is not working. So we want like every time the user always says real time that's if you say it's real time then they'll be happy even if it is internally it's near real time still they will be happy because you use the word real time.
So so technically this one is near real time because it's coming through the streaming DT and Kafka. But when we you say go and tell them we'll say near real time. So so they will hear the real time part. So then they will be happy. Oh it's real time. So now let me explain how what is the integration behind this genie like you know what's the architecture. So it starts with the teams. Then the teams have a web application.
12:53 Technical Architecture
So the teams call the Azure bot. That's the second icon. That's the Azure bot service. Then the bot service calls the web application. The web application calls the datab bricks genie API. I sounded very simple but it took us like a week to write this code. And you know the web application was written in Python. like I was in the next room in the last presentation. She was saying like you know um building it in a datab bricks apps is much easier than building it in Azure.
I agree with it because now I have to ask for somebody to create a resource group, somebody to give me a permission and then like you know the rights and everything but after it was done it worked really good and then we are pretty much happy and this is more like a archite it's a cop is a almost same as the previous one but it's like for the day like you know the developers like if somebody asks for like architectural diagram This will help.
Azure bot has lot of other channels like you can actually call that bot in Alexa or Slack or even like send like a text message. So when after you created the bot, the simplest interface calling that bot is teams because all you have to do is a simple file and then if you put it and then if you give then it will start showing up in your apps in your tenant then you have to decide who gets the rights. But Slack is little bit complex but there are some YouTubes and you know you can leverage this and this is the screenshot of our team's genie.
15:04 Genie Bot Configuration
The first one is like almost the what are all the tables materialized view we used. So it's pretty much like six or seven tables and then the second screenshot is the instructions we provided I will explain the story about that like when I go um so at the end of the presentation I will tell what to do what to not do at the time I will tell what's the problem with this kind of instructions and then the third screenshot is we train the model with certain queries so it will do its job and it'll understand and again some of the instructions is like it's an entity relationship diagram like how it's internally joined between the two uh like all those tables and this is our screenshot.
Okay, this is the interface. So now this is the funny part. I did this demo two or three times and if it works then this may be the first time it works in like a conf but it works. We promise. Yeah, I know that it works but I tested like last one hour making sure and then we even tested but time to run with scissors. Yeah. So, so I will know the answer within a second. So, okay. So, so what happens is what happens is normally you know behind the scenes it runs the SQL u serverless warehouse.
16:11 Live Demo
So sometimes we have the inactivity set to 5 minutes. So the SQL warehouse will die. So we told the user hey sometimes it's sleeping. So you have to say hi or hello to wake up. So that's what I did. So if I look go and look in the monitoring it will say hi hello how are you then it will have all those things. So now technically the warehouse started listening. So I'm going to ask some simple questions. So basically it's I asked a question like in FL99 is the warehouse name warehouse number give me top five cues but again let's see it's thinking okay good it came back I had a little bit like you know like my heart was pumping really hard okay so like bas What I did is like you know sometimes because I was training this last two hours it already knew what I'm going to ask.
Sometimes if I did not run this for like two or three days when I asked the same question list top five SKs recently the Genie team made a difference. It will ask by you want the top five based on by quantity or price or something like that. But when I came at 3:00 it asked me that question. But I said by quantity it still remembers that context. So uh I decided to take that risk instead of getting that prompt not to like you know shut it down and then my entire demo.
So here is what I'm going to do. See like um I'm going to say like um add a total row. So what I'm now telling is add a total row summing up all the quantities. Sometimes it will say if you ask the same question again and again it will say it replied back. Let's see what it happened. So, it's now like, you know, I don't know what it's going to do. See, it added a total row. So, now I'm going to say remove the description column because we are using the hotspot.
Okay. So now I can say order the and, if you asked Genie to do all of this at the beginning in your initial question, it would probably come back with that correct. Yeah. And so the idea is how can you let a user understand how this works and what kinds of questions to ask and then as people get used to using the tool now they're getting smarter and smarter how to ask these questions. So see now it's responding it's it's okay.
So I'm going to take some risks. So I this is like see if you look I have it complex. So but if it says no data return then we have to go and debug. But I I gave bunch of instruction if you list like I said list five locations for a particular skew in warehouse and then I said add a column for quantity and then I said add a purchase order number and then receiving date and then even I said oh good pretty good pretty good.
So like you know it gave all those like uh I'm actually taking more risk than I have taken that year. So I can ask what's the name of FL99 warehouse is our it's called Lakeland because we are Florida based company. So I don't know if I approved these risks. see some, you know, like like I said, it's going good. So, it's listening to me now. Okay. So, uh let me stop taking risks. Let's not do the demo anymore. Let's go back to the slides.
Here you go. So, uh these are all the references. So I gave credit to Ryan Bates and these are all the like the blogs. So and then had a very good blog and like and then is also the first three guys are all datab bricks actually the four are actually datab bricks um solution architects. So and they have all they all have a disclaimer this is not for production but it's actually in production. Okay. So that genie whatever I ran it's running in we have seven warehouses and we actually it's now running for almost 70 users and um they've been using it.
22:16 Resources and Deployment
So um in in hindsight we should have had a QR code that had all these links. If you Google Genie Teams integration these will all pop up rather than having to follow links. Okay. Let me now take you to the more disclaimer stuff. Okay. So, uh, datab bricks introduced a new connector. Okay. All the connection I did with all the like you know what they call is when I did that behind the scenes our genie is using personal access token.
23:28 Future Roadmap and Security
So if I go to my datab bricks interface, see this is the genie we are using. If I go to the monitoring tab, see it's all in my name. So like I am like one of the admin in this datab bricks. So like you know it uses I cannot figure out who is who created that qu who sent that query and everything. They made a very simple interface. This was announced in Microsoft build two weeks back and this blog explains now using Azure foundry if you from the bot if you actually connect it will use your Microsoft entra all the way up to the datab bricks.
So it will take your login id from the teams and then it will instead of using the personal access token it will use the so this is better interface and we are actually working on a version two which is going to use this n azure foundry and it's going to use the user login now we can do the role level security remember that I was still talking about the sales genie now we can have the role level security on the metalized views or whatever the view and then we can restrict and now with this connector it can see multiple genies at the same time if you have a little bit coding you can actually call multiple genies from the same interface we have a prototype but I don't have it in I don't want to take more risks like if you say like hey tickets genie ask a question it will respond Then if I say hey sales genie then it will run that API.
So and what we are doing is if I am a warehouse guy if I only have tickets genie it will then send that question to the data bricks. If not we validate what are all the groups or genies he has access based on the what groups on the groups in the entra. If I have a sales question and if I'm not authorized, it will not even send that question to uh the datab bricks. So I will suggest if someone wants to try use the idea of Ryan Bates uh blog but look at this interface and Microsoft if you scroll in the like Microsoft actually created some uh code in their GitHub.
26:26 Best Practices and QA
So this blog whatever Microsoft did I think it's take yeah see that is a blog uh and then see they have the samples it actually has sample code you can actually take it and then you can have this running in few days so uh now since we have another 12 minutes do you want to add anything or I can tell some what are all the breast practices. Oh, I was about to tell talk about the instructions. Okay. So this tickets genie whatever we did um the instructions when we did this long time back we did not we actually gave instructions how to join that's why we created that entity relationship diagram so we gave the relation like how to join but after the genie team's best practices they say if you have a primary key foreign key relation ationship between your tables or the views.
It will automatically do it because the amount of instruction we provide we actually ran out room ran out of room for this instruction. At some point of time it said we it cannot save then we called the genie team and then asked what should we do like is there is limit then they said you know if you do as a materialized view then you can create a primary key foreign key then you can remove certain instructions. So I will recommend if you are following um like I said we were one of the like early adopters so we did not know all the best practices so if you do it like like a materialized view and everything and if you have the proper primary key foreign key it's actually with that it was working even much better and it was faster.
So Nathan, do you want to talk about maybe your future plans of what is in in the road map of things that you'd like to accomplish? The future road map is first we want to have that row like the next version is going to be using the Azure foundry that's uh it's in work in progress. Then we already have a like um if I go to my genies, I already have a see this new tickets genie. So that's in already work in progress which is with the metalized view and then um the next thing is we are adding more complexity.
So now we are saying like you know empty spaces all those things and there are some cool uh announcement going to be happening in like you know they are going to add more functions to genie and that we will not we will not tell any of them today but it will be cool like um and when you meant empty spaces you mean empty spaces in your warehouse correct? Yes. Okay. Which one? like which yeah like and then calculate the um they want even more.
So even like you know uh they want to see um we call it as a transfer orders between the warehouses they want to see instead of a purchase order it should say transfer order or like you know purchase order. Oh, one of the other thing is in the regular genie interface you can ask them to create a graph but the team's interface or the API will not do it and we already requested so the data bricks team and um the idea being in teams how can a user say okay now I have this table but I want to see a bar chart showing me something else so but I don't know when it's going to come and it has a limit um 15 queries per minute.
So we actually talked to data bricks and then they up our limits so that you know but even if we tried we had like eight or nine guys simultaneously asking by the time they finish it's not one minute. So and for clarification the limit is only specific to the API not in Genie. goal is to make sure that you don't write some query that continues to ping the API and run cost all also one more thing the API will only return 500 rows so the very first day the guy asked list all the SKs in our warehouse and then it gave a data error he said your thing doesn't work then I said even if it returns all the SKs are you going to scroll like you know then he said oh no then he started so you have to be careful yeah we are out of time so thank Thank you very much.
If you want to I'm happy to if you want to ask your question here, but thank you very much for coming everyone.
Learn more about the Databricks Data and AI platform.
The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.