Sponsored by: Domo | Eliminating the Warehouse “Tail Problem”: AI-Powered Optimization

Summary

The warehouse tail problem occurs when real-world conditions diverge from static planning assumptions late in the operating day, causing bottlenecks across palletizer lanes that require constant, costly manual intervention.
Domo serves as an AI-powered BI layer on top of Databricks, continuously analyzing live data to recommend exactly what to move, where, and why—delivered through an interactive application supervisors can act on immediately.
After deploying the AI-powered optimization system, one logistics operator improved palletizer lane efficiency from 41% to 86%, replacing reactive firefighting with proactive, data-driven decision-making.

Chapters

FAQs

What is the warehouse tail problem in logistics operations?

The warehouse tail problem refers to the operational failures that occur late in a shift when real-world conditions—such as palletizers falling behind, going offline, or deviating from expected performance—diverge from static planning assumptions. These late-day bottlenecks require constant manual intervention and erode overall efficiency.

How does Domo integrate with Databricks for warehouse optimization?

Domo acts as a BI and application layer on top of Databricks, pulling live data from the lakehouse through approximately 1,500 pre-built connectors and delivering AI-driven recommendations through an interactive supervisor interface. Row-level governance ensures that different users—IT administrators and warehouse supervisors—see only the data they are authorized to access.

What AI capabilities does Domo provide for warehouse operations?

Domo's AI layer provides predictive alerts that notify warehouse teams before failures cascade, along with specific recommendations on what to move, where, and why based on real-time lane efficiency data. Supervisors can also simulate changes and trigger automated evacuation workflows directly from the Domo application.

What measurable results did the AI-powered warehouse optimization deliver?

This video reports that after deploying the AI-powered system, the logistics operator improved palletizer lane efficiency from 41% to 86% following AI-driven intervention. The system replaced reactive firefighting with proactive, continuous optimization that reduces downtime and delivers measurable ROI.

Full transcript

[00:08] My name is Mason Crane and it's awesome to be here today. Uh I am representing Domo. Just quick in the crowd, have you guys all heard of Domo? Ever heard of Domo? Can you guys everybody can hear? >> Okay, cool. Cool. All right. So, I work
[00:21] for uh Domo. I am a senior solution engineer here at Domo. And um specifically what we're going to talk about today is how Domo integrates with data bricks and how we um kind of what we do is we take data from you know all
[00:34] of the engineers, data scientists, the the really smart people doing all the cool data things and we empower the end users to be able to access that information. We kind of decouple or take away those roadblocks. We let the smart
[00:46] people keep working and we develop the business people get those insights and access. Um, specifically we're going to be operating with um a demo. Um, what is Domo? Um, and we're going to talk about the um, let me flip back a slide. The,
[01:00] um, eliminating the warehouse tail problem. Are you guys all in logistics warehousing? Is that why you're here? Did you guys just want to hear about Domo? Looking for just want to hear about it? That's totally fine. Um, so
[01:12] what is Domo? Right. So Domo is the agentic platform for intelligent enterprise work. And the way that that works is, you know, you can think about Domo as the fastest path to getting data inside of the lakehouse. Why we do that and how we do that, I really think of
[01:25] kind of Domo as five key pillars. Pillar number one is the ability to grab your data from all of its different systems via our connector library. We have about 1500 different pre-built connectors to load data into data bricks. We have our
[01:37] pillar number two is our um transformation layer where I know that a lot of you are data scientists, your SQL experts, Python experts, things like that. The where Domo comes in is we help out in that more enduser last mile
[01:50] transform type of business analytics. And then our pillar number three is our visualization layer. It's how we um it's how we start visualizing the data and and get those insights to the right people. Um the way that we do that is
[02:03] through a plethora of different options. We can do our typical dashboarding uh opportunity. We can integrate with PowerPoints, Google Slides, things like that. we can um embed our platform as a white glove approach to um however you
[02:16] want to consume it and keep it governed and then we also you know like we can email it out and schedule reports and things like that. So getting data to your end users however that might be is one of the areas that Domo is very strong. A fourth layer that we have or
[02:29] fourth pillar is our AI and workflow automations. So the cool part here is is that you've got AI integrated with data bicks and because we sit right on top of data bricks we can take all of that computing and uh AI capabilities and
[02:41] integrate it with DOMO. Some of the cool AI capabilities that we have in that regard. One is our um agent catalyst which uh is essentially like vibe coding inside of Domo using uh data bricks AI models. Um we also have our AI chat for
[02:55] business user questions and problems that they might be trying to solve. And then um like I mentioned the the workflow automations we have AI agents inside. So you can essentially create an agent for different business problems that you want to solve and then those
[03:07] agents will run kick off different processes and solve those problems. And then the fifth piece is or fifth pillar is our governance layer. So it's really just how do we keep the data secure and the cool part there is that we can integrate all the securities and
[03:20] policies that live inside of data bricks. We could also, you know, if you have single sign on and things like that, we can integrate it. Or if you're just looking for uh governance around um permissioning of the dashboards and the data sets within those dashboards, role level security, things like that, we
[03:33] have a pillar for that as well. Um now, let's see here. So, from there, I really just kind of want to um you know, help you guys understand like Domo, Domo's
[03:45] really we're here to operationalize the data. you know, you guys are doing all the hard work, prepping the data, getting it into a format that's digestible and needed for the end users, and then we are going to help get that to the right people with the right govern capabilities. Um, with all that
[03:58] being said, um, I'm going to talk to you guys a little bit about a product that I developed for, um, warehouse optimization. Okay. So, to paint the a bit of a picture for those that are not familiar with warehouse work and distribution and things like that, um
[04:12] this client operates with uh they're called palletizers and it's basically an automated machine that picks product, puts it on a lane, loads it on a truck, and gets it out the door. And what ends up happening is the upstream system that
[04:26] they have is basically operating with perfection. It says, "Hey, for a given day, we're we're going to operate 100 pallets every hour in every lane until we're out of product for the day." The problem is that upstream system doesn't
[04:38] understand that perfection is is essentially failing the moment that the day starts. Uh um a human could make a mistake, a palletizer could slow down, something could break, go offline, things like that. And those upstream
[04:50] systems, they're it it's not set up to adapt accordingly. So, this app is actually built to do that. Um, and this is just a little bit more context. I'm going to kind of click through these. Um, and so what we did is we we built
[05:04] out um some AI models. We generated an app. So this again is is highlighting Domo's capabilities of that endto-end solution. But the other cool part is is that if you guys are operating with different tool sets and you love your integration tool or your um ingestion
[05:18] tool, but you're looking for a new visualization tool, you could use Domo. Or you love your visualization tool, but you want to use an integration tool. You can use Domo as a kind of modular pillared approach. Grab what you want, use what you want. That's totally up to you. So, just for the sake of time here,
[05:32] I'm going to um pop out of this screen and I'm going to share my app. Okay, we'll bump it up into a full screen here. Okay, so what am I looking
[05:44] at? What's going on on this page? Right now we have a set of um different sites, different locations, warehouse, distribution centers essentially. And we've got insights around them. Different ones have different amounts of of palletizers. I'm just checking if you
[05:58] guys can read that. It's it's a little small, but um what it's doing right now is kind of telling you the starting state. It's telling you a bit of what's going on, the imbalancedness of the the palletizers. Um we can filter to
[06:11] different sites. We can do different things here. Um, and if I just to show you guys because this is going to work. If I come to this report section right now, you're not going to see anything. And that's that's by design because we actually haven't run anything in the model. But inside of the app, we have
[06:24] the ability to understand what the suggested moves are. We have the ability to see if you want to make a manual move. And so for those that are in warehousing or logistics or things like that, you know that there are people that have been in the industry for 20 30
[06:36] years and they they don't want data to come in and tell them how to forecast how many boxes of Cheerios do I need to order today, put it on a shelf. They don't need that. And so they want the ability to put a manual move in. And so
[06:48] we can score those moves. We can give them that capability. But what I found was really important was explaining why a move was a good move. So you'll see down here we have explanations on why they should make this move. And that helps kind of bridge the gap between not
[07:01] I don't want to say egos, but more of like the the industry knowledge that they have that they can adjust and adapt and go, you know what, I get this. Like this makes sense. I can take my hands off the off the the wheel a little bit and let this run. Additionally, we also
[07:14] have the status of the palletizer. So these can go offline. Like I mentioned, they could be working perfectly fine for a given day and then they go offline or for an hour or something like that. So being able to toggle these on and off is super important. Additionally, um we do
[07:28] have we have this AI engine. And so what this is doing is it's tracking through all the different kind of processes where Domo is integrating with data bricks. It's walking through lakehouse. It's walking through the AI query, the data sets, the client analytics, things
[07:41] like that. And so it's giving all of that highlevel tech information that you need to be aware of um for the right users. And that's another cool part about the platform. Right now as the presenter I can see everything but you
[07:53] can actually govern and maybe only the IT team is allowed to see this page. You know you can choose what content is digestible by the end users. Um so just to kind of kick this off here I'm going to go back to overview and we're going
[08:06] to notice just to call out a couple of things. We're sitting at 41% optimized. And then if we scroll down here we have our starting state and our current balance. Now I haven't done anything yet. So the current balance is staying exactly the way that it is. But if I
[08:18] come up here and I click autooptimize, what we're going to see is the model is actually kicking off different decisions and it's moving product across the different lanes or palletizers. And so while this is operating, you're seeing
[08:31] the, you know, the starting balance was from 41%. It's making all the moves and it's actually tracking all of this historical data for you so that you can actually get smarter with the model as well. And you'll get insights and I'll show you this in a moment. But so we we
[08:44] actually we wrapped it up and we went from 41% to 86% optimized. So we we did really really well. And the way that we can see that if I scroll down here and it's uh a little hard to see on the
[08:56] screen, but there are kind of the starting state was these kind of this big red bar here and now it's gray because you can see that it it it normalized. It brought the the information down. And to show this in a
[09:08] different way, if I filter to Atlanta as a site, you can see that it was starting at a starting balance of 18% and it increased to 73%. So we really normalized that data and got it to a place where you know it's going to be
[09:20] operating more efficiently. Why does that matter with these types of machines? If you're running a machine constantly over and over and over again and you're burning it out at the end of the day, it's going to wear out more more and that's going to have cost
[09:32] associated with it to maintain it, fix it, things like that. So being able to use them for all roughly the same amount of hours in a day keeps the maintenance um on you know you understand what's going on there. So now I mentioned you
[09:44] know we can come over here and if I click the reports this is showing up now and so we're seeing the amount of moves that were executed the amount of product that was moved. We can see the uh the total moves the units um rebalanced and
[09:57] then we can see a breakdown of the different facilities which ones had the biggest increase which ones had the lowest increase. And then these are actual metrics, you know, where the the the client when we were working with them, it was costing them $1,200 a day
[10:10] to not be efficient with their product. And so with um with an implementation and building this out and scaling it with Domo and Data Bricks, you know, we're we're making them or we're saving them $3.5 million a year by being more
[10:23] efficient with getting the product where it needs to go. Part of that is because you've got SLA agreements. When you put it on top of the, you know, you put it on the truck and it gets out, it's got to get there on time. you know, all those different pieces. Then you have the maintenance costs and all that jazz.
[10:36] So, if I come down here, we can see a little bit more information. We can see kind of the total moves. We can see the score distribution. And then we have some AI insights here that can help you with the upstream analytics and the downstream analytics of what you could
[10:48] do to try to potentially improve this process. And then additionally, we're tracking all the different moves. And so this is really really important because we're essentially using the app and all the existing data to create downstream data data that we can actually use to
[11:01] get better, smarter, and then you know take that all of that logic, apply it back, and then hopefully keep improving with the model um day after day. So um I actually realized I burned through this
[11:14] pretty quick. I did want to show you a couple of other cool parts. Um, so if I come up here, I'm going to reset the simulation for a moment and I want to walk through another really important piece and I touched on it briefly, but it is the um, it's the palletizer
[11:28] status. So, as an example, let's say that your your day is going great. You've got your different product, they're moving, but then this palletizer goes offline. So, what's going to happen if I click autoptimize? Now, you're
[11:40] going to see in the top there that this lane has all of this product and we have to get the product off because it's gone offline. So, what that means is, you know, this product it was scheduled to go through this lane. The lane is no
[11:53] longer operating. Let's move it, redistribute it accordingly and then see see how it optimizes after that. And you know when you have multiple warehouses, multiple palletizers, things going on offline, online all the time, being able
[12:06] to use the system to also alert end users and let them know things are going on. That's another aspect of the Domo platform where we can get those real-time analytics to your phone, um, to your tablet, to the people on the
[12:18] floor, whoever needs to consume this content, they're going to be able to get it in real time. So, if I pop out of here and I come back into the overview page, we can see that um you know, this lane it was on. It's now gone offline
[12:31] and then we redistributed as best we could based on the the situation that was going on. Um and actually we we got more optimized because we turned one off. That's surprising. Um so, let me
[12:44] see here. I mean, honestly, I I uh I joke I'm from New York. I talk very quickly. I managed to do this faster than I expected to. Um, you know, like I mentioned, uh, I'm with Domo. We're in booth 205. So, if you guys have any
[12:58] questions, I, uh, originally I didn't realize we couldn't do questions. So, um, please come over. We've got a bunch of folks that are, um, demo or prepped on this demo, can have conversations. I'm happy to answer any questions that you have. Um,
[13:12] I mean, I'm I'm really good at rambling, so like I'm not going to do that to you all. So, thank you so much for coming out and listening. Really appreciate your time today. And again, come over to 205. really excited to talk to you all. Thank you.

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.