How Domo and Databricks Work Together: AI Agents, Cross-Sell Intelligence & Warehouse Optimization

Hello everybody, I'm here to showcase how you can leverage both Databricks skills and Domo skills to develop rapidly in Domo. So Databricks has released this agent skills repository on their GitHub that allows you to automate a bunch of different things within Databricks. Domo has their own skills library and so I was able to combine the two to create a skill.

The skill creates sample data in Databricks and then leverages a cloud integration to make that data accessible in Domo without ever moving it so it never leaves your Databricks environment. And we build a front-end application on top of that data all using Claude Code and a skill.

So you can see I have this previous conversation where I said use Vessel to build a dashboard covering financials for Arches National Park incoming costs and then it goes through and within five minutes built me that sample data. It used the Databricks AI functions to generate four insights for me based off of that data and then that, you know, pretty sleek looking dashboard there. This just goes to show that if you're into building with Claude Code, you can automate not only the tasks within Databricks but also with Domo and then take advantage of the synergies between the two platforms.

Hello everybody. I wanted to introduce you to a solution that we showcased at the Databricks conference. This is a sales and marketing intelligence solution focused on cross-sell and upsell opportunities amongst our existing customer base. So to get started, I wanted to pull back the curtain and show you a little bit of how the solution was made.

You'll notice that Databricks is operating as our foundational data platform as well as being responsible for the machine learning and model serving that are so critical for this particular solution so that we can recommend the best products for our customer base. Domo is acting as the interoperability layer between Databricks and Domo, our cloud amplifier technology with read, write and native transform as well as magic ETL and model management. And Domo is also delivering the UX and activation that is necessary to be able to visualize all of this process and take action on it.

So let's jump into what the solution actually looks like. So we're going to start here with an overall analysis of our customer base. We can see what our total spend is across our customers, any active support tickets, what are the products that are most commonly being used. And I have the ability to filter on things like customer grade, industry or even just kind of manually picking a customer there.

So I'm going to go ahead and filter on all of our customers with an existing grade of B, I can now click this dropdown. These are all customers that fit that particular criteria. I'm going to click on somebody like Anderson Fitzpatrick. This shows me their monthly spend with us, any active tickets, the average usage across our platform of products, the health grade, can see the products that are being used most by them, where they're at, etcetera. A bunch of really valuable information as I'm researching this particular account.

Now where it gets really interesting is when I go into this cross-sell upsell opportunity, you'll see that it kept that particular customer Fitzpatrick and Campbell Anderson and I can now run the cross-sell upsell analysis. So as soon as I click that button, this is now communicating back and forth with Databricks running that model and soon we're going to get those results returned to us which will showcase the best opportunities for this particular customer.

So there we go. That's there. We get an executive summary on Fitzpatrick and Campbell recommending the next steps that are available and then the top three products and they're ranked, right? So Datacore Analytics is highlighted as the number one pick. It gives us a good description of why this customer is a good fit. They're in the retail industry, relatively high average revenue, indicating strong capacity to benefit from advanced analytics. It provides us a sell strategy and even key talking points.

So this is the number one thing that we're going to focus on, but it's also providing us a couple of additional options that we can consider or maybe even use as backup plans inside IQ and Secure Shield. So this is the solution in a nutshell. It's designed to leverage all components that we have from a data perspective to provide very strong suggestions on how to best serve our customers.

Hello, everybody. Today I'm going to show you our sales OPS agent built on top of Databricks. So this is an agent that aggregates all of your data from your different sources. We're bringing in data from Salesforce, turning that in Databricks, any systems that you use to record calls, we're pulling call transcripts, emails, everything. So that way, all that context for an opportunity is in one place.

We can use a summary to allow sales reps and sales leaders to know what's going on in a deal. We can use AI to identify which milestones have been met, so we're no longer relying on reps to update a CRM. We'll identify risks in the deal using AI, like so if there's any negative sentiment in a transcript, we'll flag that and then help you come up with an action plan to mitigate that risk.

One of the coolest parts is how we can hit these AI services in Mosaic to help coach on the deal prep for next call. Just have a conversation all with the context of what's going on in that specific deal.

Yeah. So right back to the CRM, any forecast comments or any other objects you want to update within your CRM so we can help you right back there as well. This just goes to show you how you can use Domo and Databricks to get more value out of your structured and unstructured sales data to, you know, convert more deals and win more deals. Thank you.

Hey everybody. So here's a quick walkthrough on the warehouse optimization app, high level understanding. This app is used in warehouses to move product inside of the building to make it more efficiently diverge across the different lanes to then be packed and then loaded onto vehicles. So the main objective is you have all of these pallets or lanes and they're distributed poorly. There's a lot of product here, low product there. We want to move things around.

So inside of the app we have the overview section, the optimizer section, which just shows all the recommended moves. We have the ability to make manual moves where you can select a palletizer and then move products to it, and then it'll give you a score. We have the status of the palletizers where we can turn them on and off to walk through fictitious scenarios. We have the AI engine walking through all the different pieces that are going on inside of it.

We have the report. This will work once I run a couple of the recommendations and then set up—just you can kind of ignore this. Don't really worry about it. At a high level, it's walking through the matrix between moves, which is how it determines a good move or a bad move. The inventory table actually shows like all the product that's on the different lanes, what's movable, what's not movable, and then an audit log of everything that's moved.

So if I come back over here and I click auto optimize, we have this visual representation of all of the product being moved across the different lanes. I think it's kind of cool. It shows where we started. So our starting balance was somewhere—it was at 40%. Then we're going to make a few different moves. And each time we make this move or make these moves, it gets a little bit better until eventually we've made all of the moves that we can make based on the logic and criteria.

So we started at 40% and then when it is all said and done—so it just finished—we got up to 85 or 86%. I think there's like a rounding issue, but so we started at 41, we ended at 86. You can see here kind of these are where they started. These gray bars are the starting lines and then how it all adjusted accordingly.

If I filter down to a particular site, it can show again, this is where we started. We were 18% optimized and now we're 73% optimized. Definition of correct or good is if the palletizer is within 20 upper or under 20% of the average. So this one came down, this one didn't quite get there, this one went up. So this is a really good adjustment.

So that's how that works. If I jump over again to the reports here. So if I go to the reports, we can see the moves that were executed, the amount of product that was moved, the average move score, total moves, some AI summary of what's going on, a breakdown per site, how many moves we made and the percent increase.

This is just—these are actual dollar amounts for the company. They're only distributing it to one facility. I made it a bit more for demoing purposes to make it a little bit more impressive, but they can scale it to other facilities that they have. They just haven't done it yet.

And then finally, we can breakdown the report. So again, the moves, the score distribution, some AI insights around what they could do upstream or downstream to improve, and then a breakdown of all the different moves and their scores that you can export if you'd like. And then again, if we go to the move log, we can see that same content over here, just a different way. So that's what's going on inside of the app. Obviously, I'm going to be at the conference. So if there's any questions or anything, just let me know and I think we'll leave it there.

All right, okay, I'm sorry. I paused the recording and then I'm thinking about it. Just one extra step. So if I go into the optimizer and I go to a palletizer and I turn the palletizer offline, it's then going to allow us to auto optimize again. This lane has gone offline and needs, we call it evacuation. So all of this product is being distributed accordingly across the lanes and then it redistributes.

Now recognizing that we started at 86% and then we went down to 84%, that happened because we obviously cannot be as cleanly distributed due to the fact that we have to get this product off. That's a requirement. So then we can show that you could then turn this pallet back on. We could re-optimize. We're starting at 84% and then we're going to start loading product into this from other places. Again, this scenario wouldn't really occur for a current batch, but it might occur for like a future batch during the day or a different day. So we started at 84 and now we're at 87. So I just wanted to walk through that extra step.

‍