Databricks Genie and AI BI: Self-Service Data Insights with Semantic Context

Summary

Genie Ontology combines two layers of business semantics: explicitly modeled stable concepts such as product lines, revenue definitions, and customer categories, plus continuously extracted inferred context — enabling any employee to get accurate, contextualized answers from natural language questions.
Traditional dashboards and conversational Genie are complementary rather than competing, with BI teams evolving from building individual dashboards on request to creating reusable data products that power both modes of analytics.
Customer examples including Rivian's rapid analytics adoption and Virtual Foundation's custom social impact Genie space demonstrate broad applicability of the Genie ontology approach across industries.

Databricks Genie and AI BI: Self-Service Data Insights with Semantic Context

Self-service analytics traditionally requires either waiting for BI teams or understanding technical details most business users lack. Databricks Genie introduces a semantic layer, an ontology, that captures both stable business concepts and continuously extracted inferred context, enabling any employee to ask natural language questions and get accurate, contextualized answers powered by your data.
Explore how Genie Ontology combines Unity Catalog's governed data with automatically extracted business semantics to democratize access across your organization. Learn why traditional dashboards and conversational AI are complementary, not competing, how BI teams evolve from building dashboards to creating reusable data products, and discover real-world customer stories including Rivian's rapid analytics adoption and Virtual Foundation's impact-driven AI applications.
🤝

Chapters

FAQs

What is Genie Ontology and how does it improve AI answer accuracy?

Genie Ontology is Databricks' semantic layer that captures two types of business context: explicitly modeled stable concepts like product lines and revenue definitions, and continuously extracted inferred context from your knowledge systems. By grounding AI answers in verified enterprise semantics, it reduces the errors that occur when LLMs interpret business terms differently than your organization uses them.

Are dashboards and Genie competing approaches to analytics?

According to Kim Wong in this video, dashboards and Genie are complementary, not competing. Dashboards remain valuable for known, recurring questions that benefit from visual context and comparison, while Genie handles ad hoc and exploratory questions that previously required waiting for a BI analyst to build a new report.

How does Genie change the role of BI teams?

Rather than spending time building individual dashboards on request, BI teams can create reusable data products — certified tables, metric views, and curated Genie spaces — that power both conversational queries and dashboard visualizations. This shifts BI practitioners from reactive report builders to proactive data product owners who multiply their impact across the organization.

What customer examples are featured in this video for Genie adoption?

Rivian is highlighted as a customer that achieved rapid analytics adoption using Genie as part of its data foundation work, accelerating self-service insights across the organization. Virtual Foundation uses a custom Genie space for social impact applications, showing that the Genie ontology approach extends well beyond traditional commercial enterprises.

Full transcript

[00:19] Welcome back to Summit Live. I'm Holly Smith and I'm here with Ari Kaplan. We have a fantastic lineup for you today. We've already had one of our guests in but now we have our second guest to talk about AI BI. You might recognize him from the big stage yesterday. Please can we welcome Kim Wong? How you
[00:36] doing? Doing great. Awesome. Busy week to you I'm guessing. It's a little busy. Okay then. fun to talk to you. Well, some people might not have seen your introduction yesterday. So do you want to give a bit of an intro about what you do at Databricks? Yeah, so I lead product for Genie, AI BI
[00:51] and our work in Unity Catalog Semantics. And I see you're wearing an AI BI piece of swag here and you've got 100k written on there. This is an old one. This is a classic. It's a classic. It's out of date. Do we need to stitch a new number on there? No, I mean Well, 100k is how many people are watching?
[01:07] Oh hey, there you go. Maybe that's what it means. But no, sorry. How are we at the moment? What? Sorry? Sorry, this is all out of date if it is. This is This is an old milestone that we hit with marking when we hit 100,000 daily active users on the AI BI. This was quite some time ago.
[01:24] Oh, okay then. All right then. So you've been able to talk about loads of new announcements this week. Yeah. What's been your highlights? I mean the most exciting thing I think generally people are excited about Genie and specifically Genie 1 and the Genie
[01:41] Ontology and the possibility that creates for democratizing access throughout their workforce. Yeah, so I think ontology ontology is something that people have been asking for for a very long time. I think there's loads of different implementations we've seen of an ontology. Can we maybe hear it from
[01:58] you? How is this different from just like a I don't know, like a network graph on top of Yeah. you know, database? So, setting aside how it's implemented in terms of whether it's a graph structure or something else, fundamentally the the difference in the Genie ontology is this belief that
[02:14] there's sort of two layers. There is the stuff that, you know, is stable in your business that you want to explicitly model out. Like at Databricks, we understand our product lines. Right. That is rapidly changing, but it's stable enough that we would want to model that out. Um the concept of
[02:29] revenue and the concept of customers. Um so that's modeled ontologies and modeled semantics. And our solution that in that space has always been Unity, right? And we continue to expand what's there. But what what um we're adding and announcing at this summit is the concept
[02:45] of Genie ontology, which really adds the idea of inferred context, which is everything else in your business that's always changing. It's very, very context and domain specific that you couldn't possibly model out. And what we're doing
[03:01] is automatically and continuously extracting that out out of all your knowledge systems so that Genie and any other agents that's connected to the ontologies is able to answer questions more accurately. Yeah. So I think in your demo yesterday, you showed a few examples where things
[03:17] didn't go so well with other agents. Um but having context is really useful. But from working with other customers on ontology, I mean, have there been any surprises about, you know, just how complex this kind of data is and being able to to model it?
[03:33] Yeah, I mean, I think the complexity is not just in the data, but it's really in the re- realization that as a decentralized data team, there is a gap between your ability to just manage the infrastructure and your ability to understand all the different ways in
[03:49] which your uh organization is using that data and what it means to them. And really that's the missing context layer and somehow you need to find a way to um enable that acquisition of semantics into your data platform without being
[04:04] sort of that middleman. Like if if every business team relies on the central data team to model out their semantics, you're just not going to get to the breadth of coverage you need to be successful with something like Genie. And also that I feel like there's a step to lose or misconstrue some information
[04:21] there. If you then have to I don't know, if you ask me to write down like exactly what does my team do so that we can represent it in some kind of agent and then give it to the context. Like I'm going to get that wrong and then it's going to be out in 6 months anyway. I think the outdated part point is a great point. Um and as I said like what
[04:37] you write down might work very well for your team, but there is many variations of your team in the organization and it's it's sort of unreasonable to try and create choke points where it has to go through through a single office. Yeah, there's like large large retailers where different
[04:54] divisions or they had different products, different acquisitions. What simple as what is a customer? What is a churn? What is a sale? Is it when you buy a product at the point of sale or what what about clothing where there's a 30-day return policy?
[05:09] So that whole, you know, unified semantic is so important to get your instance accurate. Like in that example, you can imagine a very well-run retailer would go, "Hey, I'm going to be thoughtful about this and model it all out." And cuz there is
[05:25] a implementation organizational cost and that's totally reasonable and you need to catalog semantics continue to expand to cover those type of use cases. But um if you think about an organization like Databricks, right? So fast-moving place, but think about just the engineering
[05:40] teams that we have. Every single engineering team like maybe teams of five or six or seven, they have the concept of like stories inside Jira that they're trying to figure out like how to how to implement. And they would go through some type of t-shirt sizing exercise, small, medium, large.
[05:57] Every single team does it differently. And so when you're in in in a given team asking like, "Hey, how many large stories do we have?" It's unreasonable to say, "Hey, like central data team or ask Bruce, like
[06:12] hey, define large for me." And And that will be a journey for Bruce for the next, you know, 9 months, right? By the time he gets the answer, it's going to be out of right and wrong anyway, so. true. Yeah. Uh all right then. So, uh you've probably been very busy this week talking to lots of customers. What have
[06:27] been your favorite stories that you're hearing from customers at the moment? At most right now because Genie 1 is quite new, Yeah. what I've been hearing is a lot of what they want to do. So, that's been very, very interesting to hear and it's very good to actually hear validation for
[06:43] some of the things we've done uh internally. What I've seen so far because Genie 1 is already pervasively used inside Databricks, it's like all sorts of It It's the truly the first product where like every single employee at Databricks can actually use. And so the level of engagement, like, "Hey, I I
[07:00] no longer do my monthly business reviews because I, you know, just ask Genie to fill fill it all out for me." Which is And And I think that's a great use case, right? Like obviously you create an artifact that you can further curate, but then a lot of employees are also taking that
[07:17] next step and going, "Hey, I realize now I don't need to do a monthly business review if I have an agent that can answer that business question." So, they will take that monthly business review document and say, "Hey, Genie, use the logic and use the concepts
[07:32] we've captured here to create like a business review bot, right?" And then now the executive team can go and talk to that bot to get like constant up-to-date uh um state of Yeah, and and from executives, one fun thing of being at Databricks is Ali does
[07:48] our leading of our all-hands. So, from the very top, he uses Genie on the call to do things which actually inspires everyone how easy it is to set up and to use, and then it permeates throughout the whole organization. So, you also for
[08:04] sales, like, "Hey, we're we're here at Summit. We're going to meet this company. Make a a brief. What are they using? What are they not?" Exactly. Yeah, who should we be talking to? And I will also say his demos are never like a curated and manicured. It's No, they're they're not demos. They just show you. They just They're not demos. They're not They're
[08:21] No, they're not demos. And often, like, that's that's the conversations I've been having with customers, too, is just telling them I I literally just show them, "Hey, um before I join this call, and you know, I kind of impressed them with my understanding of what's going on." And then I just show them, "Yeah, that's cuz I as I was walking here, I asked Genie,
[08:38] 'What's going on with you?'" And there's so many use cases like this. So, with all of this stuff kind of coming up with Genie and like ontology and things like that, do you think there's still space in the world for things like traditional dashboards? Oh, absolutely. Cuz I think one of the
[08:53] misconceptions is that um dashboards are a place you go to to answer all of your questions. Sometimes dashboards, and oftentimes dashboards, are used this way. It's it's to drive alignment toward a common goal. You don't want people to ask, "Hey, how's my pipeline going?" You want everyone to
[09:10] constantly think about what the state of my pipeline is. And the only way you do that is you push that into the, you know, kind of push them in the face, right? Uh and and for for that purpose, um dashboard is are phenomenal, and I don't think they'll ever disappear. Um I
[09:26] think the where dash boards fall short today is when you use it to answer all the possibilities of questions that you have, and this is when they get really bloated and unwieldy. So, I think there's always going to be a world for that. And and a lot of times I had I had the good sports background working in
[09:43] professional sports. And some humans, actually most humans, think either visually or like through the written things. Most humans still look at charts and things. But what I love about the Databricks and AI BI and that whole dashboard is it's not your traditional static dashboard where you have a team
[09:59] of people, you ask them to make this dashboard, they come back in a couple days or hours and give it to you. You can interact saying, "What does this graph tell me? Make a graph that does this." But it also could do research and it could even be predictive. Here's what our sales were, make a forecast moving
[10:15] forward. Yeah. I think that's a great analogy. Like, the idea of getting rid of dashboards is for sports, I mean. Like, imagine you're watching a basketball match, but then like you to know the score you have to ask a question. It's like it's sort of like nonsensical, right? You want to
[10:30] know the score constantly. But there might be nuances about like what is happening. Well, like did how many points did Steph Curry score? And you wouldn't want that all listed out on your scoreboard. Um and so these these things are complementary, like really not at odds. So, we probably have couple of thousand
[10:46] BI engineers watching right now. If you wanted to give some advice to people who who today maybe their bread and butter is making dashboards, how do you think that role is going to change over time? I mean, I think a lot of BI developers that I know already understand that
[11:02] their role is shifting and they're built Their job is to make sure there are great data products that are consumable for users. And the form factor of that is changing. If they now come in the form of dashboards as they traditionally have, but they also come in the form of conversational
[11:19] experience that you curate like a genie agent, right? Or they just come in a package of very well-documented data sets, which can be consumed by, you know, anything. So, I think the mindset shift of what is the fundamental job to be done it's to arm people
[11:36] with the data that they need that doesn't change for BI developers. Have you seen examples with customers where they've really taken to it like a duck to water? I mean, I'm not saying like the most kind of data savvy company but in terms of you know where they've been given these
[11:51] tools and all of a sudden it's caught on like wildfire. Like what is that kind of spark that sets that off? Yeah, I mean I think you know obviously anytime a company has a solid data foundation that's already secured and managed by Unity they they they all this stuff comes very easily
[12:08] for them. Like an example of that is Rivian. We had Mikey on stage last year and he he talks about this like all this Genie stuff for him it's all almost like a second thought because like now once he had the foundation in place all of that just comes. But what's been really cool this year is
[12:24] that as organizations are realizing the potential of what can you can do through Genie it's motivating them to kind of get some of that work done of you know hey we got to bring everything together. Okay well we have one minute left. Any you mentioned Rivian any other cool
[12:41] customer just I think over over there virtual foundation is one of probably one of the coolest stories that we have. You know they have they've been obviously they're a non-profit trying to dispatch medical services through
[12:57] analytically trying to figure out what's the most effective way and they've been working with Databricks research to develop a custom Genie which spoke to their needs. They they've been pushing us and we've and we've been working with them and I think it's an amazing story of how this
[13:14] agentic stuff can go beyond just making money and really making an impact on humanity. All right then. Okay then thank you so much for joining. I found that really useful. I hope you found it

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.