Sponsored by: Domo, Inc | Enabling AI-Powered Business Solutions w/Databricks & Domo
Summary
- Seamless integration: Domo's Databricks integration connects business users directly to Delta Lake data and AI/ML models without technical barriers.
- Performance optimization: Cloud Amplifier optimizes data processing through pushdown SQL while Domo AI Services enable leveraging ML and LLMs.
- Fraud detection demo: This video demonstrates an AI solution for fraud detection showcasing practical integration benefits.
Sponsored by: Domo, Inc | Enabling AI-Powered Business Solutions w/Databricks & Domo

Talk by Laura Qualey, Sr. Director, Partner Operations, Domo (Sponsor Speaker)
Domo's Databricks integration seamlessly connects business users to both Delta Lake data and AI/ML models, eliminating technical barriers while maximizing performance. Domo's Cloud Amplifier optimizes data processing through pushdown SQL, while the Domo AI Services layer enables anyone to leverage both traditional ML and large language models directly from Domo. During this video, we’ll explore an AI solution around fraud detection to demonstrate the power of leveraging Domo on Databricks.
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
FAQs
How does Domo connect to Databricks Data and AI platform?
Domo's integration seamlessly connects business users to Delta Lake data and AI/ML models while eliminating technical barriers.
What is Cloud Amplifier?
Cloud Amplifier is Domo's optimization layer that pushes SQL queries down to Databricks, allowing complex processing at the data source.
What is AI-powered business intelligence?
AI-powered business intelligence combines traditional analytics capabilities with machine learning and generative AI to help organizations derive insights and take action automatically. Rather than requiring manual report creation or static dashboards, AI-powered BI systems can discover patterns, predict outcomes, and recommend actions based on business data without explicit programming. This approach democratizes data-driven decision-making by allowing business users to get answers through natural language queries instead of waiting for analysts to write custom reports.
What AI tools let business users build agents?
Domo provides tools that empower business users to build AI agents through its integration with Databricks' Genie and other AI capabilities, enabling non-technical teams to create intelligent workflows. Unlike traditional BI tools requiring SQL knowledge, Domo's interface allows business users to define data questions and actions in natural language, with AI handling the technical translation to queries. The platform handles authentication, data access, and integration with backend systems, allowing business teams to build custom AI-powered solutions without data engineering involvement.
What are the best AI-powered business intelligence tools?
Domo combined with Databricks represents a leading solution for AI-powered business intelligence, offering real-time data access, advanced analytics capabilities, and seamless integration with ML models trained on Databricks. The platform's strength lies in connecting business users directly to data and AI without requiring technical intermediaries, while maintaining governance through Databricks' Unity Catalog. Other competitive solutions may offer specific features, but the Domo-Databricks combination provides the most comprehensive stack for modern enterprise BI.
Full transcript
Okay, perfect. Well, let's get started. We've only got 20 minutes, so we're going to jump right in. So, today we're talking about enabling AI powered solutions with data bricks and DOMO. I I work in our technology partnerships group. I've actually been at Domo about 10 years, which hopefully right now you're thinking she's too young to have been anywhere 10 years. Thank you. I'm flattered. Uh Braxton Fullen Camp is also with us.
He's going to be walking us through a live demo here in a bit. So for agenda, we're going to go through introduction and what Domo is, how Domo works with data bricks, and then we'll be highlighting our fraud detection AI solution and kind of just going over our AI capabilities and how you would build that in DOMO. So first off, I guess I'd love to have a raise of hands if you've never heard of Domo before this week.
Okay, I got a couple. Yes. So thank you for coming to listen to us. Um, we've actually been around for a while, about 15 years, but we kind of focused more on the analytics component in those early years. Um, now we introduce ourselves as an AI and data products platform. And what that means is we allow you to connect to data, prepare smarter insights, and build agents faster. So that seamless data integration, that's dynamic business intelligence and the firepower for those with workflows and automation.
All of that, of course, built on a robust govern framework that allows you to have tight security and clear visibility into your entire data ecosystem. Okay, I want to talk briefly about analytic stack. And if you're not familiar with that term, it's really just the life cycle of your analytics. So kind of left to right is like maybe you're using Fiverr or Stitch on data ingestion. So you're bringing your data in from your sources cloud data lakeink onrem and then you might be pushing that into an ETL or ELT layer such as Matilian or Informatica and then eventually moving that data into data bricks so that you can build your models and store on top of your data lake.
Then we also have on the right side kind of additional tools like Tableau and PowerBI that might be grabbing the data from data bricks to visualize for your business users. Um, additionally, we're also seeing customers who also add OpenAI co-pilot or hugging face in as additional A IML capabilities. And what Domo does is we actually provide a unified analytics stack that allows you to have a single scalable and fast solution all in one place.
So, we can layer into these components if you have multiple tools. So, kind of left to right, we have an fantastic self-s served data acquisition layer. So, that's a thousand plus connectors. We can connect to anything Then we'll push that data for you into data bricks. You can layer on our magic ETL that's kind of a canvas experience drag and drop tool for configuring data transformation um domo AI services. And then as we move to the right side of the architecture those are those end you know data consumer experiences that you create for business users to be able to access the data.
So that's dashboards that's low code no code procode app even apps that you can generate with AI. And we're going to talk today about agent catalyst. And agent catalyst is really our umbrella term for our AI toolkit. So those are all kind of the agentic components that you need to build agents in your environment. And I want I like to think about AI solutions as kind of a pyramid. So the AI agent might be at the very top.
But in order to get to that, you have to have the foundational components in place to get there. So that's you know at the very base layer on the left side of our architecture here. Do you have your data silos broken down? Do you are you uh you know unifying your data in one location for those services and tools to grab? Is it clean? Is it transformed? Do you have sustainable and scalable data architecture and pipelines?
Then if we move up a layer, it's the tool set. You know, does does your AI agent AI agent have access to the tools it needs to actually action on your behalf to automate to make decisions? Then above that, let's say you've got those two in place, you're able to make an AI solution. There's a question that comes up a lot at Domo which is what makes an AI solution agentic and I like to think about this in two ways. I feel like an AI solution becomes agent if it can reason on its own based on the context knowledge and information that we've given it and two it can h have access to tools to actually take action.
It can self-improve. So with that context one last bit on the data bricks component. So you're able to create and maintain tables and data bricks using demo connectors our magic ETL to build data pipelines. So that works by you know we have data transformations that are pushed into data bricks as optimized SQL and then we're going to spend a lot of time today on the slash components that's building smarter apps and AI agents on data bricks.
So that's leveraging work you've already done in data bricks models um to deploy and take action on them in DOMO. And that brings us here. So I'll let Braxton introduce himself. He's a principal solutions consultant at Domo with a wealth of knowledge around AI and how Domo works. He handles our most complex customer use cases. So if you want to talk to him about what you're trying to solve at Domo, he's a great resource for that and I'll let you take away with the live demo.
Awesome. Thanks. Can I get a thumbs up if you guys can hear me, too? All right, cool. I just want to make sure we're switching over. Um, we're going to walk through a live demo today. The demo today is focused on this credit card fraud transaction use case, but as you heard say, Domo can solve really any AI use case through the tools that we have, especially as we wrap around the data bricks platform. So, uh, this is sort of the end view of what we're going to see today.
We're we're going to leverage some machine learning models in data bricks. We're going to use leverage data that's in data bricks and then we're going to leverage the Domo tool set on top of that to um access those machine learning models. We're going to leverage AI in the form of an AI chat experience. So we'll get back to this dashboard in a minute. I want to walk you guys through the journey of how we built this solution.
Starting off with our data warehouse. You can see here I'm actually connected into multiple different data bricks instances. We support, you know, a multicloud strategy and having access to these multiple environments. To create this integration, we're going to use our cloud amplifier configuration here and we're going to register a new integration with data bricks. All you need is uh your connection URL and your personal access token and you're off to the races.
We can configure this integration at three different levels. The first one is the connect or the query only integration. So that's where we can leverage data that's in your data bricks environment and we can query that natively in Domo and start to build visualizations in in really really rapid time right in in just a few seconds. The next level of integration that we can configure is this native transform and that is using Domo's magic ETL tool which is a drag and drop business userfriendly ETL tool to transform that data but with all of the data processing happening in your data bricks environment.
So we're going to use our our ETL tool. We'll show it to you here in a minute. And all of that processing is being pushed down into data bricks. So none of the data is leaving the environment or or you know even coming into DOMO. And then the final level of integration that we can build is our write integration. And this is the ability to use Domo's connector library of over a thousand pre-built API based connections to hydrate your environment to you know put this data into your your data catalog and make it available.
And so often times customers will use you know the endtoend suite to build solutions but we also have customers that just leverage DOMO for an ingestion or a transformation or just the visualization. So while it is a a wide array of capabilities that you have you can choose you can pick and choose which ones you'd like to use to solve the problems. So in our use case we have data that's already in our Domo environment.
We've done the read and the transform and the write integrations. We've brought this data in and then we actually passed this over to our machine learning team or our data science team and they've built us a light gradient boosted model for fraud detection. So in this case we're just leveraging a model. You can actually build these models in Domo. We have integrations into Jupyter Workspaces and other areas that you could develop these in Domo.
But in our case, we just developed it in data bricks and then we're able to reference that model or really any model that you build and deploy and host in data bricks. So in our case, we've added our model here. We've chosen data bricks as our adapter. We've gone through selected our model and we've registered that in Domo. And that makes this available in multiple areas of the platform as well as in a a standard uh API call interface.
So you can access all of these different models leveraging the same user interface. So we have our uh gradient boosted model here. It has an endpoint. We can test this model. So this would be an example of doing live inference. We've got our base data schema here. We've input and in this case this transaction came back as zero. So this was a nonfraud transaction. Now this is great that we can call this interactively and we could call this interactively through an app.
But the more likely use case is that we would want to run this in batch or we would want to run this as these transactions are coming in. And so now we're in our magic ETL interface. This is a drag and drop interface. We can bring in multiple data sets. We have utilities like adding formulas running SQL natively in in data bricks value mappers. So it's really meant to be for both your low code and your high code resources um and associates who would be leveraging this.
In our case, we're taking this schema and we're flattening it to prepare it to run inference. So, we're joining all these things together. You can see this is very point-and-click. I'm doing an inner join here. We support full inner, outer, left, right, cross join uh with equality conditions as well as expression conditions. So, if you're doing uh you know a join between dates, things like that, that can all be configured here using our no code or low code interface.
And this tile here is our AI service tile and it's our AI model inference. So we're leveraging that same model that we saw earlier here and we're picking the task and essentially at this stage in the data flow we have the data prepared in exactly the schema that the uh service is expecting and then we're passing this in and looping through to create uh a final output data set that is going to have our classification.
So each transaction is going to be classified as a zero or a one. And we're also going to pull back some metadata from the model and uh you know have that data uh available to the downstream steps. And so the results of this this data set is now an object in data bricks but it's automatically registered into DOMO as uh a data set. And so we can build our analytic solution. What we're looking at here is an app in DOMO. apps in Domo are both visual in nature such as a a dashboard.
So you can see here it's fully interactive. We can switch our our um metric here from revenue to transaction amount and we'll see our charts update. We can zoom in on a a specific time period and we're going to see that everything on the dashboard is now updated to that filter. So fully interactive experience. This is all built using no code. This is our our drag and drop experience. Each element is called a card. Those are built in a tool that we call analyzer.
And again, that's very business user friendly. No SQL required, anything like that. Uh we can also do some what if analysis. So this can be interactive, not just swapping out dimensions and measures, but also as I adjust our fraud probability here, I can see our prediction distribution. I can also drill into each one of these transactions. So as I mentioned in our machine learning model, we were doing some Shappley values on that.
So I can drill into a specific transaction ID for a specific customer and I can drill in and now I'm getting Shappley prediction values on each one. So not only do I know that this is a high likelihood of fraud, but I can also identify why was that chosen as a high candidate for fraud. So I can can see these different values and that might help me determine what I want to do to mitigate this in the future. Alongside the dashboard, we also have our AI chat experience.
This can be powered by any large language model you would like. You will register those in that same model management interface that we saw earlier. So if you want to power this based on uh you know a model hosted in data bricks or you guys want to use open AAI or or or any um model you can leverage that here and we can go ahead and um ask questions which transaction subcategory has the highest risk for fraud. We are then uh transcribing that using a large language model into SQL.
We're then running that SQL against your data bricks environment, pulling back those results. We're going to build a chart for you. Then we're also going to give you a natural language response for that you can copy out. You can then adjust the chart type if that makes sense. Um you can continue on. This is very conversational. So if you want to ask, you know, additional clarifying questions, you can do all of that in the AI chat experience as well.
You can also save this visualization and go edit it using that no code analyzer experience as well. On top of the analytic experience in the app that we've built here, we also have an agent running in the background. So every time we run that magic ETL interface and we come up with those new inferences, we're taking the deltas, the new data, and we're passing it through our agent. What we're looking at here is called a a workflow in Domo.
So again very visually similar to our magic ETL but I call this our our magic ETL but for real life we can take action on the data beyond just you know processing data we can u take action on it and the first action that we're taking here is for each new transaction that comes in that's flagged as fraud we are running it through an agent an agent takes four ingredients the first ingredient is the input so in this case I'm passing in a specific transaction my transaction ID the flags This is the information that's coming off of the machine learning model.
I'm then giving my agent access to tools. In this case, my agent can query a data set with SQL. So, it can query into that transaction level detail information. It can query information about the account, recent transactions, things like that. I'm giving it access to the AI metadata. So we can either integrate with your data catalog or we can allow you to input metadata about your different columns such as column descriptions or synonyms and that information is going to help the agent understand how to query your data set.
So based on the information that it's seeing in the transaction, it'll break it down. And then we're also giving it the ability and you can see we can provide subcriptions here or or specific instructions for these tools. We're saying, "Hey, we want to send stylized emails to an an email address based on the outcome of this." So, we have our input. We have our tools. Our next step is our instructions. Here, we're providing the agent with a specific set of instructions.
So, we're saying, "Your job is to analyze each transactions fraud indicators. Um, compare the probability to the thresholds that we've, you know, preset." So, kind of a rules-based engine enhanced by this AI agent here. And then we are asking it to make a decision on which path we should take. So what is our output? And so our fourth ingredient here is we're configuring our output. What do we want to come out of this agent?
And so we've got our agent here. We're getting our decision. And then we're choosing a path here. We're we're letting the agent choose the path. So are we just going to notify someone via email? Are we going to escalate this to another team? are we going to flag that transaction and assign that to an agent to actually go be a human in the loop to review or in in some cases we might actually choose to block some transactions so we can use API calls from Domo out to external systems and do things like block transactions in real time and then we're logging all of those decisions back to a data set that is going to be um available in data bricks so we can use that for downstream analytics we can use it in our dashboard etc.
Reason I call this an app and not just a dashboard because most of what you've seen here is a dashboard is we also have the ability to create these interfaces by which humans can take actions in Domo. And so our app here is going to come up and it's going to have our list of transactions and will allow us to investigate finite details about that and also take action. So if one of those steps that we have in the workflow is to have a human in the loop, we can do that using this uh this interface once the Wi-Fi catches up here.
So now we can see each transaction ID here. I can drill in. I can see information about that transaction. I can have uh a link out here to the approval queue. So I can go over to our our Q's product and approve this or or deny it as you know actual fraud and uh and things like that. So, thank you again for your time today and I'll pass it. Yes, thank you so much. So, we are around. If you've seen our booth over there, I need to do a plug for Bougie Bingo.
We are giving away some insane prizes. At 2:30, we're giving away a $1,200 travel voucher and at 3:30 a Nintendo Switch, too. So, please stop by the booth. Also, if you have an idea for an agent, we'd love to hear from you. We can slot into any tech stack that you've got going on. If you have any um data issues that you'd love to talk with us about, please stop by. Thank you so much.
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.