How to Turn Every Business User Into a Decision Maker With AI/BI and Genie

Summary

Unity Catalog Metric Views create a governed semantic layer that defines KPIs once and shares them consistently across sales, finance, and operations teams, eliminating the inconsistent-metrics problem where every team has a different number for the same business question.
AI/BI dashboards can be built with drag-and-drop tools or AI-assisted chart generation, and Genie Code demonstrated live building a complete multi-page executive dashboard from a single natural-language prompt in approximately 10 minutes.
Genie agent mode cuts a root-cause analysis that traditionally takes 7 hours of analyst effort down to 15 minutes by routing natural-language questions across multiple Genie agents and connected data sources without requiring any SQL knowledge.

How to Turn Every Business User Into a Decision Maker With AI/BI and Genie

This is a practical session for leaders who want to move beyond static dashboards and operationalize data-driven decisions at scale. Youssef Mrini and Josue Bogran show how Unity Catalog Metric Views, AI/BI dashboards and Genie replace fragmented reports and spreadsheet chaos with trusted, conversational analytics. See how governed KPIs create a shared semantic layer so sales, finance and operations speak the same language while still moving fast. Through live demos, follow a business user from the Genie homepage to a dashboard, then asking “why is this changing?” in natural language and exploring scenarios on top of trusted metrics, no SQL required. Leave with a blueprint to turn report consumers into proactive decision makers and close the gap between insight and action.
Talk By: Josue Bogran, VP of Data + AI Architecture, zeb.co ; Youssef Mrini, Sr DevRel, Databricks ;

Chapters

FAQs

What is a Unity Catalog Metric View and how does it work?

A Unity Catalog Metric View is a governed semantic layer built natively into Unity Catalog on the Databricks Data and AI platform that defines KPIs once and makes them available consistently for both AI and BI use cases. It eliminates the common problem where sales, finance, and operations each calculate the same metric differently, resulting in conflicting numbers that slow decision-making. Metric Views can be created through a UI builder, YAML configuration, or Genie Code, and they serve as the single source of truth for dashboards, Genie agents, and downstream analytics.

How does Genie agent mode work for root-cause analysis?

Genie agent mode allows business users to ask natural-language questions — such as 'why is this metric changing?' — directly against governed KPIs and connected data without writing SQL. This video shows it reducing a root-cause analysis from 7 hours of analyst effort to approximately 15 minutes by routing the question across multiple Genie agents that each specialize in different data domains. The result turns report consumers into proactive decision makers who can investigate anomalies independently.

Can Databricks replace my existing BI tool like Power BI or Tableau?

This video presents AI/BI dashboards on the Databricks Data and AI platform as a complete BI solution, showing how they can be assembled using drag-and-drop chart building or AI-assisted generation without requiring dedicated engineering work. Genie Code is demonstrated building a multi-page executive dashboard from a single natural-language prompt in about 10 minutes, which the presenters position as a direct alternative to maintaining separate BI licensing fees. The platform also supports scheduled reports and alerts for always-on delivery to stakeholders.

What is Genie 1 and how is it different from a regular Genie Space?

Genie 1 is described in this video as a single unified interface that routes user questions across all Genie agents and connected applications, so users do not need to know which specific Genie Space or data source to query — they simply ask a question and Genie 1 dispatches it appropriately. It is also available as a mobile app on iOS and Android, enabling data-driven decisions from anywhere. This moves analytics from a desktop-only, report-consumer experience to a conversational, always-on interface for every business user.

Full transcript

[00:08] My name is Jose. I'm the VP of data and AI architecture at SE which is a consulting company. I'm also part of the product advisory board with data bricks data bricks MVP. I advise a handful of companies too. Uh gotten a chance to
[00:20] interview mate reel a lot of the founders. With me I have the one and only Yousef. Go ahead introduce yourself. >> Well you already introduced me. My name is Yousef. I'm a Deil here at Datab
[00:32] Bricks and I've been with the company for the past five years. >> Awesome. Well, today that's a very long title. Uh so I'm going to skip saying that because I'm probably going to stumble, but what I will say is we're
[00:44] going to show you how data bricks does BI and AI with BI. Okay. So let's start off by asking a question. I think that's on the next slide, I think. >> Yeah.
[00:56] >> Just so I understand the audience a little bit. How many of you would consider yourselves more towards the analytical type? Okay, that that's kind of what I expected. Okay, so next fun question.
[01:09] How many of you are PowerBI users today? I'm I'll add myself to that. How many of you are Tableau users? I'm not going to raise my hand on that
[01:22] one, but they're they're great tools. But how many of you if data bricks offered something that was a complete solution for your BI needs, how many of you would
[01:36] like to still keep on play on paying for your licensing fees? So that's what that's that's the vision behind some of the things that we're we're going to talk about today. So with
[01:51] that said, uh if you can show the agenda, please.
[02:03] There we go. Uh, quick agenda for today. I'm going to talk a little bit about the history of AI and BI inside of data bricks and I'm going to tell you some fun stories around that and then Yousef is
[02:15] going to do a little bit more of the technical deep dive uh, doing a few different things. But like I said, we want to make things engaging. And while at first we're going to show you some
[02:27] slides, we're going to do things a little bit different. We're going to start with Yousef dropping in a prompt on Genie Code. That's going to take like maybe 10 minutes to run. And then, you
[02:42] know, you already see he already started it right now live. and then hopefully in 10 minutes from now it's going to be ready for the next part of the presentation. So we wanted to that's our last minute change for you. Um you go
[02:55] ahead into the next one. So like I said I like I like to talk a little bit about history. I've been in the data bricks ecosystem for like five or six years. Got to help behind the scenes with a lot of the different products.
[03:08] So, data bricks started having a BI or dashboarding type solution I think like 2015 or something like that and it was inside of notebooks and I'm curious has
[03:22] did anyone ever use dashboards inside of notebooks with data bricks here? Did you like them? >> Did you like them? >> Okay, I'm glad I wasn't a customer back
[03:34] then. So, I I skipped that part. uh with all respect for the folks that were that that did do that. Um then eventually came Lake View dashboards. Okay. And I may
[03:48] have skipped another product, but you know, renames happen all the time with data bricks. Um leg dashboards had a little bit more a little bit more power to it, but it
[04:00] just still wasn't quite there realistically speaking. Um I two years back quite literally two years ago I messaged Ali Gutsy and I said hey
[04:15] data bricks does a great job around data engineering and also AI or back then when it was cool to say ML ML right and I said you guys do a great job with
[04:28] that but when it comes to your dashboarding solution Is it even a priority for you? Thought that was a great question. So um
[04:41] especially when in the time you know within the context data bricks was more and more competing with like the fabric. So this was his response. His response was you can read it yourself but it was basically like
[04:54] within 12 months it's going to look a lot better. It's going to be good uh 24 months from now. So exactly two years ago, it's well two years and two months
[05:06] ago. You're going to be basically amazed with uh what you know the whole dashboards family of data bricks. And he also said that Genie that was part of the vision for Genie and dashboards to
[05:18] merge together eventually which we're more and more closer to where that is at today. He basically said it's going to be a great thing. So I I'm not going to
[05:30] be the judge whether the dashboards or genie is amazing or not. That's going to be you hopefully after this conference trying things out and seeing what works or what doesn't work for you.
[05:42] But hopefully today with what we're going to cover, Yousef is going to demonstrate just what you can do. And again, you go back to your business and you try it out and really vote with your money except not with your licensing
[05:55] money because you don't you don't have to pay licenses licensing fees for this. So with that said, Yousef uh next slide please. Uh he's going to show us how you can go from raw tables and I'm not
[06:08] talking raw tables as in unrefined data but you know tables that are the data looks nice and clean or fairly clean. and he's going to show us how to go from from that to actually having governed
[06:23] metrics. Take it away, boss. I'm happy. So, hopefully after this session, you're going to all be power users. And let's all start with think this is something
[06:37] important. I'm pretty sure you must have at least once asked a question and depending on the team you're targeting, you had a different answer. And this is
[06:50] the main example I faced so many times like what are the number of active customers across the company. If you ask the uh finance, they will say okay it's
[07:03] 184,000 customers. Why? because it's the number of built customers for the last 13 days. If you ask the marketing team, they will they will share with you another number based on maybe the number of emails opened. If you ask the product
[07:17] team, it will be based on the weekly active users. And of course, if you ask execs, the numbers always rounded because it should look perfect. >> I I I don't think that's ever happened to them. It's just the organizations you've worked at, man. So, never
[07:29] happens. >>
[07:41] is it the the marketing is it the finance the product and that's why people started talking about semantic layer and trust me they solve a huge problem because they this
[07:54] is the let's say the the bridge between the technical folks and the business users it's going to translate or the uh like the KPI in complex SQL queries
[08:08] but again it did create another issue because most of the semantic layers at are at the BI level or the dashboard level and this is problematic because if I am a data scientist I want to access
[08:20] those KPIs well I can't then I need to build again those KPIs and I will be facing the same problem same thing for the engineers and also about governance. How am I supposed to make sure that the KPIs that I'm building are only visible
[08:34] by by those who needs them? Especially if I'm I don't know, I'm a public company and I need to make sure that the quarterly like uh numbers are only displayed to like the boards. So I need
[08:47] to make sure that everything is governed within the platform. >> But I mean I I have a question for you everyone here. How many of you are being pushed internally to do AI related work?
[09:02] I work for a consulting company. So I got to say that um what happens to your business when your business logic for AI lives in a different place than the one
[09:15] for all your dashboards. That's for me on personally that's been one of the biggest realizations for me when I realized like why is data bricks pushing so much for like the BI stuff. It has been that realization of like oh
[09:28] because guess you're going to have again different answers when so much of that work is also so much of those worlds are blending in together. So that's why >> I like to interrupt him all the time so
[09:40] I apologize >> I got used to it. Yeah. And that's why data bricks did introduce Unity catalog native semantic layer. Basically you can define the KPI once at the catalog level
[09:53] which means whether you use like no matter which BI tool or your dashboarding tool you've been using you can access those KPIs at least a huge bunch of them and what's cool is that
[10:05] those KPIs can be accessed through the like dashboards notebook AI agent trust me this is very important specifically we're building agents for literally every single task and of course it's
[10:17] governed with unity catalog. This is very very important. And you're going to all ask me a question. Okay, that's cool. But I'm a business user. How am I supposed to create or build those uh
[10:29] this semantic layer? Well, I have the answer for you. But before, let me show you why it's powerful. Imagine I'm asking this question. What's the average revenue per customer per segment for f
[10:41] for it's hard for fulfilled ordered since the first January of 2024? I have two answers over here. So I have the SQL query and trust me it's a little bit complicated for BIS for business
[10:54] users because it's like a lot of joins orders count discount aggregation not null repeated logic you more likely going to make a mistake and either even within with an AI without context you
[11:06] will more likely have a a wrong query and the same query with UC semantic layer it's way easier you know that you're not you're going to select the customer I'm going to get the revenue
[11:19] new per customer and I'm going to or I'm going to uh like filter by the order category valid that I asked there and now you can you you're able to read the query in much easier and by the way that
[11:31] I choose the red the the orange because because my friends always wear a red hat >> it's just coincidence that I have all this stuff on so >> maybe sucks I don't know >> and we have three ways to build uh to
[11:45] build the semantic layer on datab bricks so the first one is the UI builder and I will show it to you in few. Basically, you do drag and drop. You just select your tables and you start building your KPIs with drag and drop. It's so easy
[11:57] and it's can be used by analysts and domain owners. Then the second options are YAML files. So this is mainly spec as code for those who loves being techy and geek and you know you can deploy
[12:11] through CI/CD and dab data bricks asset bundles or they change the name and it's best for data engineers and platform and then we have the easy pass that's
[12:24] genot well you can do it just by describe it in English I mentioned here English but it can be French or any languages or sp Spanish Spanish or so on and you just write prompt and Genie will
[12:36] build the the semantic layer for you. And that's what I and it can be done by anyone. Now I I've got to I got to do a poll here. If your favorite method being mentioned here is the UI builder, can you raise
[12:50] your hand please? If your favorite measure I mean your favorite method is the genie code, raise your hand please. if your favorite method is Jaml.
[13:03] In all honesty, that's a lot more on Jaml than I expected. So that I won't make the joke that I was gonna make anymore. So, >> and remember I did a prompt. Well, you can see I use Genica to build my
[13:15] semantic layer. So, basically here uh I will share with you the prompt if you want, but I'm saying we have a data set on data bricks called bake house for a fake uh bakery in New York I think or in
[13:27] many many cities. And here I'm saying here hey those are the sources you sales transaction sales franchise customer supplies and I want you to of course join the data and those are the those
[13:39] are the um joining key and then I want you to build uh some dimensions transaction day transaction month build measures out of revenue quantity transaction and so on. So we can see and
[13:52] I see like current day trailing for like last seven days 14 days uh and then add the day overday percentages. So we see it did build my uh semantic layer over here. So we can see the sources the
[14:04] different tables I have over here my measure my my source. So basically I did everything just with uh just with uh genie code and just to go back to the to
[14:16] the presentation and we will deep dive a little bit later on it. So I did mention three key pillars. I mentioned the sources. So basically the sources those are of course the views the tables that
[14:30] you will be using to build your semantic layer and of course you're going to have many tables. You need to join them. So make sure to define the right uh the right keys. This is this very important. You got to select the dimensions. So
[14:42] basically dimensions are the columns that you'll be using to aggregate your your your metrics which is for example can be uh c customer it can be product it can be country city so you can have
[14:55] all of them there and then you have the measures and the measures this is the one of the most important part this is how you're going to define how you going to compute the KPI for example which formula are you going to use maybe I
[15:09] don't know revenue divided by 12 or revenue plus commissions. You know, you can have so many uh measures. And this is something something that I find very very interesting. Imagine you did
[15:21] compute a measure. Let's call it uh I don't know uh it can be uh like yearly revenue. You can then compute another measure by referencing a measure you already computed. So for example, you
[15:33] can say uh like monthly revenue is yearly revenue divided by 12. Again, it's a wrong formula, but just saying that you can reference a measure in another measure, which going to makes like the the way of reading those
[15:45] measure way easier than having to read a complex SQL uh SQL query.
[15:57] layer, but how am I supposed to use this semantic layer? Because I need to use it somewhere to make sure that I can consume my assets. Well, that's where you can build AIBI dashboards that Jose
[16:10] uh talked about and again we have many many options. The first one is build with the UI and again trust me I'm not good when it comes to building dashboards but I am now an expert thanks
[16:23] to well not this one second option and how you can do it. So basically you can just select like it's a drag and drop you select the visualization you drag and drop it then you you can add filters
[16:36] and you can also use AI assisted which means you can just for example select a path and write down like revenue per product or per per month. It will automatically build the chart for you and you can of course build chart after
[16:48] chart using the AI assisted and within few minutes you would have built your entire dashboard just with AI and of course any everything built with AI can be edited in the u in the dashboard uh
[17:00] with the UI of course it's best for analysts and business users and you can also use the second uh yeah I should explain this also something important I think especially this one because I
[17:12] already explained the others uh it runs directly on the lakehouse, you no longer need to duplicate the data. You no longer need to again duplicate it and run another governance on the top of it. You already always access the same data
[17:24] with the same governance layer. No longer need to repeat everything you've been doing and maybe avoid using Excel for some reason. And >> happy compliance team
[17:36] >> if you're in a regulated industry. >> And we have also some interesting features there especially for salespeople. You have forecasts integrated directly in the platform in the in in the UI which means you just need to select the time time the time
[17:48] series and the the other interesting column that you need and then it will automatically run the forecasting for you on the dashboard. You no longer need to ask the data the data scientist to build it for you and beg to make it
[18:01] available test it. So you you're now a power user. That's what I've been saying since from the beginning. And of course the other easy option and that's why I told you I'm the master of the dashboards now just ask me to not to not
[18:14] do them manually you I'm I can use genic code. So basically that's what I will do in few minutes in front of you. Again you can write that by using natural language prompt. It will automatically
[18:26] generates visualization for you. It can build many charts, build many pages and of course it can and everything you're going to build with the uh with gen ed editable through the uh the UI
[18:41] and and trust me this is super super interesting and I I will show you an example. I will run it. I will let it run because it takes few minutes and then I'll show it to you in a few. So
[18:54] >> while he does that, how many of you here use chat GPT cloud code anything like that?
[19:07] >> Um how many of you ask a question direct ask a question about your data or your business directly into Chad GPT or Claude and you don't try to audit the results.
[19:22] So the reason why I say this what Yousef is about to show the value I think is clearly evident that when you have um you know those those models connected to governed data
[19:36] the quality and how reliable all those answers that you're getting just shoots up through the roof. I don't know if everyone saw today Ken Wong present at the in the keynote when he said how like
[19:49] you know they had the thing had hallucinated like 24 customers or how or 24 people present at the p thought that was very funny and with Genie well I would say always check the output of anything you're you're seeing um it just
[20:04] makes your life a lot easier and you can you can trust the data trust but verify always >> what I have the answer for you as usual. But before I know that
[20:18] dashboards answer questions like for what like what was my last week sales what's my best product that's good you can find this easily but when it comes to why understanding what drive what did
[20:32] drive the sales to go up or go down understanding from where come the outliers then you need to request your analyst hey I want you to understand what's going on my sales are going up or maybe more likely when when they're
[20:45] going down, I want to understand what's going on. This takes time because it requires the analyst to open many tables, run multiple queries. It can take maybe days or weeks depending how
[20:58] big or how large are your your data set and your of course the number of tables and how complex is your question and we're able to answer those questions. And trust me, I did run it. I did run
[21:10] this. I asked one data analyst to debug a real issue. It took the analyst seven hours. Genic agent mode was able to solve it in 15 minutes. And you can try
[21:23] it uh on your end. If you have access to the platform, do it. If not, you can try it on by creating an account in free edition, dropping maybe not not like real data, maybe fake data. It's pretty
[21:35] important because it's another platform. We're talking about the free the for clarity. We're talking about the the free edition, not not the one that you use for your business. So you can trust you can trust that one. >> Just don't use real data. Use fake data
[21:48] and ask some just to give it a try to understand how it works or you can use it uh on your platform. And basically how it works and sorry I think the screenshot doesn't work doesn't display very well but basically the agent you're
[22:02] going to ask a question what are the outliers? So what will happen because unity catalog has access to your the metadata the context the lineage it will create a plan by by like and this plan
[22:14] will will be based on hypothesis like you can have maybe 10 15 questions asked by model and then the SQL warehouse will run those will run those those uh
[22:26] queries and compare the result uh reflect on them and then it's going to build a report for you and then you can just Check the report. You can download it as PDF and share it with your shared it with your leadership. Trust me, it's
[22:40] so powerful. I trusted I I did it with so many uh customers and it gave them so many inserts, so many insight in couple of minutes. And before I think we can go
[22:53] back to the dashboard or not yet, it's not there yet. And the the the most interesting part that you can turn Genie into an always on analyst. Imagine being
[23:06] able to ask question anywhere at any time. And how you can do this? Well, you can do it through Genuan. It used to be called previously data bricks one for those who may have
[23:19] seen it before, but now it's called Genuan. And basically it's one experience, zero friction. Basically it's uh it's uh you can open it through the browser you login with your SSO and
[23:31] then you're able to consume all the asset that has been built by the analyst data engineers it can be dashboard it can be genaces and it can also be data bricks apps
[23:44] and and in case you have maybe I don't know 15 genaces because there is one for marketing one for finance one for sales you can still ask a question a generic question and we have uh an agent that's
[23:57] going to understand your question and redirect it to the right gen space in case you don't know which one you need to target and then it will of course use the agent mode to answer your questions
[24:09] and give you insight to go to be as as I mentioned a power uh user and how and one of the most interesting features that I really advise you to try is this schedule and alert run so what you
[24:22] imagine you're again I of like being uh sales director of a of a sales company. So basically um I can schedule a query over here by asking a simple question.
[24:37] How was my sales last week and what's going on? So basically I know I have dashboard that can tell me hey last week you had I don't know 100,000 sales. That's great but I want to understand is
[24:49] it going up is it going down and what's causing them? Why did it happen? >> Why did it happen? >> Everyone can Everyone can query. Everyone can export to Excel and run some quick numbers. But digging into why it's happening, that's the part that
[25:02] took your analyst a while to figure out. >> Exactly. Sometimes it can be it's too hot, so we sold so many juice. Can be it's a holidays, Super Bowl, so many chips or whatever going to eat.
[25:16] I'm not American, but I'm trying not to >> never would have guessed that. So you can do it, you can schedule it and every mo every Monday morning you're going to get the report. So you no longer need to bother the the analyst to
[25:29] have you do have those questions. You can be the super agent again. You can be the power user. And of course you can access everything I mentioned from your mobile phone. I
[25:42] wanted to show you mobile mobile. Yeah, it's here from from your mobile phone. Just download the Genie uh Genie one mobile like uh Apple iOS and Android
[25:55] >> and on Android. >> Yeah, on Android on both platforms and you can and also you can reference them in Slack and Teams in case you have them. And basically you can ask questions at any time, literally
[26:07] anytime. You have access to the dashboard, you have access to Genie and everything. So basically imagine like you woke up at 6:00 a.m. in the morning. You want to know what was your like sales last week and understand what's going on. You can just use Genie. You
[26:20] can just use data bricks. Genie one. I was so confused. And now hopefully my uh my prompt is done. >> We have a plan B if it's not, but hopefully it is. >> Oh, it's almost there. So you can see
[26:34] that I have my executive summary, my sales performance, but it's still running. uh franchise customer analytics. So you see it's uh it's still running. It takes time but just to uh
[26:47] show you that it I really have one. It's just takes a little while. So it's here eight days ago. So this is still going to run. So basically this is the the
[27:00] results. Oh, it's slowing. Here we go. So we have the results. Let me let me open all of those. >> So we have the executive summary. So we see the trailing I mentioned for the past seven days, 14 days, months. I have
[27:15] pie chart. I have trends. I can see again the sales performance. Uh the franchise like uh by by uh by transaction volume by by country, the
[27:27] customer analytics, the daily trends, uh product inventory and the customer sentiment. And basically it was done with a simple uh with a simple prompt saying like build a comp. Yeah. Yeah.
[27:40] Like you see it's currently populated. Build a comprehensive multi-pace analytical dashboard like executive summary sales performance franchise. Imagine what you could accomplish with the basic prompt and you can do more.
[27:54] >> You can refine what you're doing. For example, I would never have a donut chart on my presentation. I would tell Genie there like, "Hey, get rid of that doughnut chart. Give me something else." So you can be very specific. Give it this title. Give me this kind of shard.
[28:07] Do legends in this way or another. Uh for me I come from an analyst background. Originally made my way back to data engineering. I I I love this. I love this. And you can also access Genuine. So this is as I mentioned the
[28:21] the consumer part of data bricks dedicated to business users. So as you can see you have the dashboard. So all the dashboard that has been built for uh for uh for you. So you have all the
[28:33] Jenny spaces that are available. You have >> the apps >> never mind it is never >> you have the you have the apps and you can also have the the schedule that I did uh that I did mention
[28:46] this piece. Uh maybe it's not me that created this problem. So basically let me go to okay because I didn't create genpace makes
[28:58] sense. So let's create the first genace. So I'm going to select the KPI that I did mention over here. I'm going to build the uh so here we go. We have my genace and then I can ask a
[29:10] question. Analyze the outliers. So this is what I asked the question. Remember I mentioned when I use when I'm
[29:23] using the agent mode I have a plan. So the the agent mode will create a plan based on the hypothesis. they will run many queries then going to understand things that are how things are going on
[29:35] then going to share with me a report. So that's why for those kind of questions it makes it it makes it can take times and what's new you can click on the tinken and you can see the different questions that the system will try to to
[29:48] ask for example you can see here top revenue outlier by transaction entity franchise outliers during zcore analysis product outlier top revenue outlier by transaction franchise outlier product you can see you can see how the agent is
[30:01] thinking and those are literally the same questions that an analyst will try to solve in order to share with you the results and basically you will see that it will it will keep asking questions
[30:14] understanding how it works and after a few minutes you will have a basic report shared with you that you can download as uh as a PDF I think it will take time so we going to start taking questions and
[30:28] hopefully it will >> first of I want to ask a quick question did how was that helpful to everyone here >> awesome that's

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.