The Descartes Systems Group Inc. ($DSG)

Hi, everybody. I'm Bob Bowman, Editor and Chief of SupplyChainBrain. Welcome to the special presentation, The AI Exchange Inside the Last Mile AI, delivery, engagement and the new standard for on-time and in full. First to 4 webinars on the theme of the AI Exchange to be presented by Descartes this year. Quick reminder folks, there will be [Operator Instructions] So what if your roots could learn from every delivery in this AI Exchange session today, we're going to explore how moving beyond static service time assumptions leads to a new level of fleet performance. Traditional routing treats each stop as predictable, but in reality, each 1 is shaped by order size, product mix, site conditions, unloading requirements and crew readiness. AI and machine learning changed the model by learning from actual delivery behavior and continuously applying that intelligence to future routes. So let's see what our experts have to say about it. I'd like to introduce our speakers for today. Sergio Torres is Senior Vice President of Product Management with Descartes. He's responsible for managing strategy and vision for the company's entire portfolio of routing, mobile and telematics solutions. Prior to joining Descartes, Sergio worked as a Director of Business Development and consulting in Europe for [ CAPS Logistics ]. And Cyndi Brandt, who's Vice President, Fleet Solutions with Descartes. Cyndi has deep expertise across product management, product marketing, sales enablement solutions, engineering and marketing [indiscernible]. At Descartes, she leads strategic efforts to bridge customer needs, with innovative solutions for last and final mile operations. Welcome to both of you. Welcome audience. With that, I'd like to turn it over to Cyndi for a brief presentation.

Thank you Bob for a great introduction for both of us. Descartes were all about technology that moves over. We're a global leader in logistics technology that really helps over about 26,000 customers completely around the world, help get things from point A to point B, and we want to help do that in a smarter [indiscernible] and more efficient way. So for me, it doesn't matter whether it's raw materials side either for a factory or finished products that land at your door, our technologies behind the scenes, making all of that happen seamlessly. But it's a very, very [indiscernible] at all. We are a talented team of 2,200 plus pros who love, love, love solving tough problems and working side by side with our customers to help them succeed. And then lastly, as a company, we've had about 20 straight years of record performance, and we interestingly reinvest about 15% of our revenue back into R&D because we want to keep that innovation flight. . In our particular pillar, which is Fleet Solutions, we have a delivery and performance management platform. It optimizes wholesale distributors to optimize routes, streamline dispatch, execute daily operations with precision and confidence. And we do that while making sure that there's real-time driver and customer engagement happening out in the field as well. We want to make sure that we're capturing valuable feedback, fuel continuous improvement, and power that improvement through AI. So it's a great topic for us to talk about today. Then we look at it is every route, every driver, every delivery. It's optimized, connected and built to exceed expectations.

Thank you, Cyndi, and thank you, Bob, for that wonderful introduction. So artificial intelligence at the [indiscernible], this security presentation of what it is. But Look, it starts with a very simple idea, using data to make smarter decisions and automate complex logistics workings. So at the broadest level here in this picture, you can see AI is the overall field. So within that, we have machine learning that focuses more on systems that learn from data, improve over time. So this is where we start to see the real operational impact, especially in areas like route optimization, real-time optimization and continuously improve ETA predictions. We'll be talking about more than this during this session. So going deeper, deep learning uses neural networks to handle more complex patterns. This enables capabilities like OCR, image recognition and predictive modeling, turning on structured data into a global insights. On top of that, you have generative AI that introduces a new layer of value. It can create content, summarize information for us, and power conversational experience. Descartes, this shows up in things like technification, customer insights and chatbot interactions that enhance our user engagement. And finally, you see at the very bottom, agenetic AI, which represents the next evolution. These are autonomous systems that don't just analyze, they actually observe, plan, and act. The idea here is to we will be orchestrating multiple tools and leveraging Gen AI that they can execute workflows end-to-end with minimal human intervention. And as you can see here, the power of this is basically the data that we are collecting. So combining all these comprehensive data, name it, route planning, route execution again. Drive -- how the drivers are actually doing when they are delivering a service to your customers, driver safety or driver behavior when they're actually driving a truck and we want to know if they are actually doing harsh-braking or speeding, solid acceleration. All this data combined in something we call the fleet date intelligence, actually allows us to provide data insights to our customers, provide analytics and also provide Gen AI agents. Now this is important because the Gen AI agent is what makes it actually the smart part of it. It actually looks into insight, summarizes it actually suggests and also it provides the ability to execute tasks. So we're kind of closing the loop and we call that [ Rene ] agent. So from fleet intelligence, basically getting all that data close in the look to make improvements. So think about this example. If you're actually receiving your planning and your execution data, and you want to see how your plan compares to your actual execution data, Rene will be looking into your data to figure out where are the opportunities for -- to improve efficiencies and to improve the utilization of your fleet. So Rene will not only say, "Hey, by the way, it looks like you're actually overestimating your service time durations. You should probably look into that. Would you like me to do that for you? And that will basically enable Rene to activate machine learning for service time predictions and that is executing an automated task for you. And at the same time, to be able to say, no, I want you to utilize these service time predictions at the moment of the optimization of a route. And that will trigger the configuration of your route planning solution to start consuming the service predictions, and that is closed in the loop and little by little, closing that gap between planning and execution.

Thank you very much, Sergio, and bringing Cyndi back at this point as well. We now move into the panel discussion portion of our presentation in which I have the privilege of asking Sergio and Cyndi to help us dive deeper into this whole question of AI Audience do remember that we're going to be leaving room at the end for an audience Q&A, so please do be submitting your questions all along by clicking on that Q&A button at the bottom of your screen.

So let's start out here, guys. Let's start with kind of a discussion of AI in real world delivery. Cyndi, AI is becoming embedded in day-to-day delivery operations, not just long-term strategy. What do you see as the most immediate challenges that fleet leaders face today? And how have those pressures evolved over, say, past 2 or 3 years as expectations are on time and in full delivery continue to rise.

Sure. I mean that's a really interesting question because what we're seeing right now is that AI is really landing kind of right in the middle of some very, very real operational pressures, right, not just these long-term transformational goals. People can't wait that kind of have a real impact. But when I look at whatever fleet leaders dealing with today is just really a tough combination, right, rising costs, unplanned rising costs in an exponentially fast right. A lot of labor challenges right now. Customer expectations continue to change every single day. But we see over the last couple of years and even in the last couple of months, right, fuel, insurance, equipment, wages, everything's increased. So these inefficiencies now are really starting to double up. Some little things that might have gone unnoticed, they're now really getting your margins almost immediately. And at the same time, expectations around delivery have completely changed again, right? 2 or 3 years ago, customers were okay with a nice broad time window today because of the B2B expectations -- or B2C expectations have been set rather, they expect precise super accurate DTAs, a lot of real-time visibility to what's happening with their order and a seamless experience, right? And again, they wanted to mirror that home delivery they're getting from a retailer. I think the biggest true shift is -- it's how last mile delivery is starting to be perceived, right? It's no longer just a function, get the trucks out the door, make the customers happy, right, not just the simple logistics function. It's become a mission-critical part of the customer experience as well as the cost structure of the business. It's forcing different fleet operators to really try to strike a better balance, if you will, between the efficiency and service quality than they've ever had to look at before. .

You remember, 2 or 3 years ago, we said things like customer demand is greater than ever before. Little did we know where it was going to be like today. So you really got to step up and AI is just absolutely essential to that capability, obviously. But let me ask you to drill down a little bit further, Cyndi. Many fleets they're still experiencing a 10% to 20% gap between plan routes and actual execution. Again, this real world kind of thing. Where are these breakdowns happening most often? And what does that mean for service reliability and customer experience?

Sure. I mean, that's still a pretty big gap, right? I mean we can all do better [indiscernible], but we do consistently see that 10%, 20%. To us, it really comes down to the planning system is still operating on assumptions. The real world is more dynamic and far more precise, right? The breakdown really tends to happen when there's a disconnect between what was planned based on all those assumptions, right, those educated guesses and what actually unfolds in real time on the road. Think about traffic variability, service time inconsistencies, service time estimates, block docs, even last-minute order changes or driver behaviors. They all introduce friction into that deliberate process that you can never really fully capture and plan especially when you're making generalized assumptions. Fleets have to start to feed execution data. So what actually happened, [indiscernible] into that planning process. If they don't, those same inaccuracies are going to continue day after day. Routes might look optimal on a piece of paper, but they really don't reflect those real world conditions, and that's why you consistently see that 10% to 20% gap. When we dig into what this really means for service reliability becomes super significant. The plan ungrounded in reality, routes aren't realistic, ETAs become less accurate, on-time performance starts to suffer. Your customers start to feel it, and that just makes your dispatch teams so your employees and start to become firefighters and feel that pain as well throughout the day. Mistime windows, poor communication and consistent delivery experience. These are absolute no-gos for customers these days. there's an opportunity here, right? There's an opportunity to use AI to really make an impact on this and close that loop and pulled the data in so that you can continually learn from that execution piece, and make your planning more adaptive and realistic over a longer time set.

2 Okay? So a real need for better planning and operational precision. Sergio, with that in mind, these traditional routing models, they rely on static service time assumptions, always have up to this point. So why has that approach persisted, however? And how does it fall short when it's applied to real world again there is a real-world delivery conditions like job site variability, product mix and unloading requirements.

Yes. static service times really persists because it's very simple and easy to put in operation. I mean if you just take averages and fit them into your day integration into your routing system, that's a simple need to actually get everything going, right? The problem is that your plans may not really reflect the reality. So if every stop is assumed to take roughly the same time. It's very straightforward to build routes and standardize your planning. But the problem is that, that worked reasonably well when delivering environments where there is no complexity. However, our deliveries are not complexity, right? We all know that. And so today, that simplicity on the data really becomes a limitation. And we have always said it in systems, Garbage in, garbage out. So the quality of data is super important. Now where you get that data is also super important. So Think of this, if a delivery to a retail location is fundamentally different from a construction side, why would you use the same service times or for a multi-drop in a commercial center, right? So when those differences really are not captured, you're actually introducing a systematic error into every single route. And what happens is that you're going to have very low compliance from your drivers to actually execute the routes that you plan. So over time, this will be leading to miss-delivery windows or inefficient use of your fleet, and a lot of mid-day adjustments. So you're actually always reacting to what is really happening in -- with your drivers. And sometimes you really have to be almost like a fire fighter looking into every single route to see where you can actually avoid any -- or propagate more exceptions. So look, well, these models are convenient, they really no longer match the complexity of modern delivery operations because now we have data that we can take advantage of that we before we didn't have. We were just working on statistics. Now we can learn from it. That's where the AI and machine learning comes into play.

Okay. So what are the results though. We're talking here about fleet shifting to machine-learn service time predictions. They continuously learn from actual delivery durations. So again, real world. What changes in planning accuracy. And how does that enable improvements like up to, what, 30% greater route density without adding trucks or drivers. Sergio?

Yes, roughly. That's -- I mean, that's a great question because, look, the -- you'll see that's kind of the low-hanging fruit. Let's look at our service times and how they are actually represented in our planning system so that we can actually give routes that are actually feasible or they actually are close to what the reality is. So when fleets are actually shifting to a machine learned service time predictions based on the actual data, the biggest change is planned accuracy basically is grounded more in a real-world behavior. So now you have a closing relationship between planning and execution. So instead of relying on the static assumptions now the model will continually be learning about what the driver is actually doing when he's delivering, delivering the [indiscernible] goods. And that data factors many variables like in [indiscernible] you're talking about order size, you mentioned it earlier, prototypes, customer location, geographies, equipment requirements. And you know the same historical patterns that you have every time you deliver to a given customer. So that comes into play as well. So that allows us to produce routes that are much closer to what actually happens in execution. So you will see very quickly that drivers stay on schedule more consistently. And then your dispatchers, your people that are actually managing operations, we'll spend less time reacting to issues and overall operations will become more stable. So -- and importantly, very importantly, the accuracy -- these accurate actually translate into better utilization. That's where the 30% comes into place. And in some cases, improving route density significantly without adding any trucks or drivers. You see that right away. The data insights and the machine learning will tell you, a, actually, you're actually closer to your execution and you will be probably saving time or maybe because you remain -- your drivers are going to be closer to the SLAs that you have promised to your customers. So it's both a service improvement and an efficiency gain in my opinion.

And you're making life better for your drivers, too. What about the human element of that? I mean, obviously, the customers are experiencing better service. You're doing more efficient operations from year-end. But all of this running around and firefighting could drive these drivers crazy. And so it sounds like this is really helping in the way of making the driving experience better, especially at a time when it's so hard to find good drivers. So yes, thanks for that. Okay. One of the big thing, one of the big truths about this whole world, you said many, many times, you cannot manage what you cannot measure. Cyndi, there's an awful lot of noise around AI these days. But where are you seeing a measurable impact today in delivery operations, whether that is fewer missed delivery windows, reduced idle time, improved asset utilization like we were talking about. Tell me about that.

Bob, I think it just is my favorite term, which is there's so much noise in the market about AI. And most of it's -- people talk a lot about AI-based routing, what they're really talking about is AI-based execution, where I'm going to move stops around as the route transpires. They're really not talking about using AI and planning. At Descartes, we've taken a different position and saying, "Hey, look, we've got to close the gap in a way that's unique and different". And we think the gap exists today, again, between the planning and execution, we've out this both Sergio and I earlier, right? When you apply AI to the execution data, what really happened throughout the day, things like your actual route service times, driver behavior, it becomes much more grounded in reality. And if it's the plan that's more grounded in reality than the execution and comparing the execution to the plan becomes more interesting, right? So if I have a bad plan and I compare execution to it, I'm not showing where I can really improve because I had a bad plan. [indiscernible] open a site plan and I can show what I did I now have the ability to isolate where I can make those improvements, which leads to here's how you get to the very measurable, if you will, outcomes. . So first, a couple of reality things, right? More accurate ETAs, more accurate communications, less missed delivery windows, less extended at run time, less over time. These are things that are very clear that we can execute on from a metric standpoint. But we shouldn't let this kind of real-time decision through the day of operations get away either. Dispatches teams are constantly [indiscernible] manual disruptions, right? AI can dramatically adjust the routes to do what's safe based on what's actually happened, right, traffic delays exceptions, I note that the dock door has been blocked. But we're going to try to reduce the idle time, so the wait time of drivers, improve that route adherence and keep drivers just moving along the road more efficiently throughout the day. We also see a lot of measurable gains in the categories of asset labor utilization, right? When I learn from that historical real-time data, I can better sequence stops, reduce the unnecessary miles. It makes smarter use of the available capacity, whether it's vehicles from human beads. And this becomes really important because of cost pressure, right? How do I use my people and might put on it more efficiently? And then just to put a bow on it and tie it all together, the customer experience is also going to improve. So your customer satisfaction rates, your customer retention rates. When they get better ETAs, less surprises and better communication, magical things happen and it becomes a real competitive advantage. So it's not 1 or 2 metrics. There's a broad swap of metrics, but take work to get there.

And Sergio, service times, they can vary by 20% to 40% based on factors like customer type, order size, geography, vehicle constraints and the like. So how does machine learning account for that variability in a way that static models cannot. And how does that translate into more predictable delivery performance?

Yes. It's very interesting. From all the data we have analyzed how much discrepancy you can have by having the static service times. And really, you come to realize that the service times are not depending only on 1 or 2 factors. They actually can be variable. So this is where it becomes really relevant. So look, service time variability is probably one of the biggest challenges in any delivery operation. And you see that because of the percentages that you were just mentioning, Bob. And it's exactly where we can actually use machine learning to understand. What are the factors that are actually affecting that delivery time? Like we said, we were talking about prototypes, we were talking about order sizes. We're talking about the location that you're delivering. Again, the accessibility that you may have to that location? And if you're visiting that customer constantly and you realize that, that customer always gives you extra -- basically introduce extra service, you need to take that into account and the machine learning will learn from that. Even instructions that you may have on a shipment will definitely affect your service time. And the beauty of artificial intelligence is not like you actually prescribe and say, this is with the way you have to actually predict service times. The beauty of it is that it's actually learning based on the context of what is actually happening in operations with the actual documentation and data that you're sending it into our routing system. So this -- all this combined is basically allowing us to have a prediction of service time that reflects more accurately and account for variability across our different types of deliveries that you're executing. So the net-net is, look, the result is a plan will -- a plan that has a building uncertainty in it, basically will lead you to better predictability execution. And this will basically reduce your downstream disruptions. Why? Because now your people are actually going to be working on a more proactive mode rather than a reactive mode. And that's the whole idea. And it unleashes a lot of benefits. Like we said, one of them, one of them is basically a better utilization of your fleet. Your drivers, as we mentioned earlier, they're going to be happier because they're actually getting routes that are actually realistic. Instead of getting the routes that they may not realistic, you see drivers coming back to the depot talking to their dispatchers. I can do this route in more time because you're actually underestimating the service times or vice versa, I can do it in a faster way. And now your SLAs that you're actually committed to your customers will be actually better met, behaving better predictable routes that look more like in execution. So machine learning is definitely making a difference in this case, Bob.

Thanks for helping you to understand the whole concept of machine learning versus AI, that could be very difficult to parse those different terms. But I guess we're learning that real world, the term real world equals unpredictability. They're almost interchangeable words, are they not? I want to talk to you about the whole concept of connected operations because so many of this stuff has been fragmented up to this point. Sergio, many fleets are still operating across disconnected systems for routing, the dispatch, for customer communication. What does a more connected data-driven delivery environment look like? And how does that improve decision-making across the day?

Yes, that's a very good question. I'm actually privileged to have participate on these products that we have at the [indiscernible], it allows us to basically have that fully integrated solution footprint, and that -- that's called that this division between system is nonexistent at also. So a connected delivery environment is really about bringing, obviously, your planning, which before it was actually separate, your execution that again, it used to be separate and the communication into a single operational flow, communication to your customers as well. So when something happens here, you basically need to make sure that those 3 systems stay connected and the propagation of data works seamlessly. So look, again, historically, they used to be separate. They were managing different systems. We used to have routing in one place and dispatching and other, and then customer communication somewhere else. And you're probably sending data between systems that is probably stale or it's actually old or it's not real time. So that really creates a lot of gaps in visibility. And it slows down the decision-making that you need to have in the real-time environment. So when these systems are integrated, now the data is actually flowing seamlessly across the operation, fleets can see what is actually happening in real time and respond quickly to any exceptions. So that level of connectivity basically allows them to manage the entire delivery process for -- again, and I mentioned this earlier, from a more proactively and consistently manner. And the fact that you have this data flowing consistently and being able to react -- to send proactively notifications to your customers if there is a delay or they could be basically -- you actually may be actually arriving earlier this actually improves the efficiency of your fleet, avoiding any delays or avoiding any missed deliveries because the customer is not available, et cetera, et cetera. It's just a cascade of benefit that comes with it.

I'm glad you brought up the idea of responding because a lot of this conversation up to this point has been about how you can better predict what's going to happen, but not the most sophisticated AI model in the world, let alone the most intelligent human being, can 100% predict what is actually going to happen because real world is unpredictable, real world, there's a certain element of chaos. Let's face it. You have to be able to respond accordingly. So Sergio, as plans inevitably change in the process of execution, how does combining planning data with real-time visibility let fleets adjust routes dynamically, while maintaining service commitments? Sergio, I can't hear you all of a sudden. Are you I think we've lost your. No. Cyndi, can we hear you?

I'm here.

Okay. You sound like you're muted, but I don't see a mute form. I don't see what we could do here real quick or we can turn to Cyndi for that if we can't hear Sergio anymore. Sergio, you might try muting and unmuting on your audio there to see if that brings it back.

All right. How about now? Can you hear me.

There we go. There we go.

Probably my audio switch to something SP1 Yes.

No,did I say about unpredictability. Okay, folks. Here we go. A real-world demonstration of what we're talking about today. So Again, let's talk about the whole idea of response, real-time visibility, planning data combining with that, respond to that.

Thank you, Bob. Yes. So look, in any delivery operation, we know plans will be changing the day, right? That's just unavoidable. This is the nature of the beast. The key is how we effectively respond to those changes. So when you see planning the data that is connected to real-time foot facility, fleets can actually make informed adjustments as conditions are changing, right? So that might mean rerouting or updating ETAs, as we mentioned earlier, or even reallocating resources to work on any exceptions. So for example, if you most resequence a route to minimize the risk of missing time windows, you must know where your drivers are at any point in time. You don't want to be resequencing and driving them crazy when they are probably en route to a delivery. And then at that point, you can decide what and what cannot be changed. And the changes that you're actually making to your delivery plan have to be communicated obviously to your drivers in real time and also to your customers to avoid any future exceptions. So having that planning data connected to the real-time visibility is it enables you to make a decade an effective systematic, systematic decisions. So instead of reacting after service levels are impacted, okay, they can make control adjustments that help you maintain the performance that you're expecting. That's the real shift. And it's from -- like I said, it's actually for moving from this reactor firefighting to a more proactive -- and it's important, data-driven decision management. This is where the power of data coming from planning, coming from execution is actually collaborating to make sure that you have [indiscernible] your fingertips in real time so that you can actually do proactive decisions and educated decisions.

Yet another truth. It's all about the data. So we hear that in any technology initiative and especially when it comes to AI applications and the like. So I want to talk now though, about what it really is all about, and that is the customer experience. Cyndi, once the route is in motion, delivery becomes the customer experience. So how are leading fleets using real-time visibility and predictive insights to manage the full delivery journey from dispatch through final confirmation rather than just through a series of individual stops.

You're right, Bob. It's not a series of individual stops anymore than it was for a very long time. It is really a continuous customer journey, full of expectations, the continued change that we talked about earlier, right? What this looks like in real practice, though, is leveraging a customer experience or a customer engagement platform. This platform has to work in conjunction with your plan and real-time execution. So there's 2 pieces to this. It's pulling in information about the plan, but also pulling in live route data and tracking progress throughout the day, right? And then it's going to give a predictive insights to understand not only where is my driver, but how is the rest of that day actually unlikely to unfold. So that might be sent in a notification instead of saying, I plan to be there at 2:00 and sending a notification at 1:50 PM that says I'm 10 minutes away. Well, now I'm not going to arrive until 02 :05 PM. So I might ask on that notification to 1:35 PM with that 10 minutes there, right? So it's being a little bit more precise there. When I look at how do companies create a competitive advantage today, it really is communication via customer notifications, right? That's the way to create that competitive advantage. It's really the crux of everything. If you're trying to find that advantage where margins continue to get smaller as costs increase, you have to have to provide kind of these dynamic event-driven communications around the entire order journey from the minute I've taken the order to the moment that it's delivered to the final moment where I provide feedback on that delivery or that driver. Leaders in the space are going to be adapting technology that's not just where is my truck, but where's my truck with context.

Right. Every time you make a communication with the customer, you are, in fact, raising a customer expectation to meet that promise of that communication. So you better be right about what you're telling them, what time you're going to show up. You're going to have some very angry customers. Sergio, I apologize if I'm sounding a little bit about this whole concept of response because that is really what we're talking about here today, more even more of the prediction of how you actually respond to the, again, to the real world. Sorry about keep saying that. But as plans change throughout the day, and of course, they do. How does connecting routing, execution and customer communication in 1 platform help fleets to manage exceptions in real time while still protecting service commitments.

Yes. I mean again, this is where the pressure comes. I mean you guys have all used over and the ability to actually see real time what is happening with the driver when it's coming to you, allows you to make decisions as a passenger. Even if you -- for whatever reason you need to [indiscernible] locations or maybe you don't want it anymore, you can actually hdo that in real time. And that's exactly what we're talking about here because exceptions are definitely going to happen no matter what, in every delivery operations, talk -- you name it, delays changes, on expected conditions, new orders coming in, new pickups that you need to do and you need to collect. These type of things basically change your plan. But the good news is if you have real-time communication end-to-end from the planning all the way to execution and to the end customer it allows you to actually be more efficient and advise people ahead of time to actually remove that noise and avoid any inefficiency that actually may actually be generated from it. So the difference now is the ability to manage those exceptions in real time. That's the reality. So when route execution again and the communication to the customer are fully connected, fleets can quickly adjust to changes, update ETAs and notify our customers if there is any exceptions like, hey, I'm probably going to be we're going to be 30 minutes late from the ETA. And if the customer basically tells you, hey, I'm not going to be there, I don't even show up then you know that immediately you have to adjust your plan or vice versa. I'd like actually now to move my delivery to the afternoon because we're not going to be pressing that day. That actually -- that information that comes back. And the moment that the customer actually does that. And if it is connected to your planning and execution, systematically, you can make a decision that is based on the data that you have received up to that in real time. So this is just basically a way of minimizing the impact on any disruptions or delays, missing deliveries, not shows, et cetera. So it does help us maintain service commitments even when things don't really go as expected.

It's funny because we talk about how demanding customers are, but at the same time, they're kind of forgiving if you level with them. If you tell them what's really happening. If you don't, that the worst thing you can do is not communicate, have something show up way and you do tell them it was going to be late. That -- speaking as -- we've all been customers in that area, and we don't like it. So it's a good point to make. Okay. Cyndi, we were just talking about what's happened with our customer expectations in the last 2 or 3 years. So the point now where on-time delivery is no longer enough. Customers also want full transparency and coordination. So how do capabilities like dynamic ETAs, estimated time of arrival, proactive notifications like you were saying, and centralized communication, how do these things reduce friction, improved job site planning and enable faster service recovery when issues actually do arise.

No. Like you said, customer expectations really and truly have evolved because we have so much access to data, right? It's gone from simply being" on time" to a fully informed, fully coordinated and predictable experience, right? We want to have the same experience every time. I want to know all the information that I can about that delivery piece. But if I can coordinate all this, that's where all these capabilities can really start to make a significant and meaningful difference within organizations. When I think about dynamic ETAs and productive notifications and centralized communication, fleets have the ability now to kind of stop reacting to managing that entire delivery journey in real time. In the past, we get a phone call, where is my truck, and we would just immediately drop everything and react to make phone calls. Now I'm pulling all the friction out of that by allowing the data to work for us, right? So if something changes super early in the route, I'm getting data information in about that. The system can start to really anticipate what's that downstream impact. There was a track down. It slowed things down by 10 minutes. Nobody is going to be upset if your ETA is impacted by 10 minutes as long as you know about it. Adjusted ETAs notify those customers, I'd like to say, trigger operational interventions before they happen or escalate within there. By reducing the friction and with all this transparency, right? When you think about job site coordination, specifically like freeing up people to meet the truck, ensure you have enough labor to meet the truck, making sure that your workers get to a building site to meet the truck that has a tremendous amount of expensive materials on it, right? I can plan and anticipate better. And customers know when to expect deliveries, they can make hadjustments to their day so that they can take those deliveries? Because remember, if they don't have the right people putting around. Well, first of all, it's expensive that people waiting around for delivered. But if they don't have the right people, ultimately, you as the delivery company is going to be delayed, and it's going to impact other customers down the road. The last thing I would say to you is that it doesn't really stop the delivery itself. When you connect a proof of delivery with feedback capture, you're also closing that loop, and then you turn your execution piece, not only into insight for planning, but insight into your customer that feeds continuous improvement, right? Does [indiscernible] restaurant always project half of the fresh vegetables that are delivered, right? There's all kinds of things that you can start to look at to create insights, to create insights into ordering patterns. That's a whole another way to apply AI. But at the end of the day, let's work on creating a controlled, consistent customer experience with fewer surprises for those customers, less firefighting internally for our dispatch and operations team and a delivery journey that's really more managed as opposed to kind of going back to that 1 stop at a time.

No. Okay. Sergio, I do want to quickly touch on this, is thing that any time technology comes into play and you have humans using it. The systems become more intelligent. But how do you ensure that planners, dispatches and drivers actually trust the outputs and use them consistently in day-to-day operations. .

Yes. Look, a successful adoption comes down to making this part of how the operation runs every day. It's not just about -- technology is an enabler, Bob, and humans make it happen if we actually run in every single day during our operations activities. So the fleets in my opinion, that see the most value, they really don't treat these type of things as a technology project. It should be aligned with operational objectives, goals, such as -- I mean Cyndi has been talking about it in her responses, like improving on-time performance, increasing route efficiency enhancing your customer experience. And that's how we want to see it. So we're going to be using machine learning for those purposes. And also, we have to involve the people that are going to be using on a daily basis, such as your planners, your dispatchers, your drivers, feedback is super important, right? It's as long as you actually involve them early in the process, the system will fit naturally into their workflow. So when that happens, technology really becomes a tool for better decision maker, it's just, it's just not just another system you're going to be actually managing. It's actually giving you the freedom and the knowledge and the foundation to actually base your decisions on and be more effective when this happens. And it actually creates confidence at the same time. So it creates confidence on your planners, dispatchers and drivers, but also on your customer base as well that you know that you're going to be compliant with their SLAs that you have promised in the past. So that's where the real successful adoption comes from.

We're short on time, but I just want to get a quick answer from you Cyndi. Over the next, over the next 12 to 8 months, what capabilities will define high-performing delivery fleets.

On the planning side, let's get away from static optimization. Let's get the plans that are informed by historical execution data and become more dynamic, right? Customer engagement, you've got to fully integrate that process into what you're doing. It can't just be notifications. It's really got to be information about the entire order journey. If you bring these 2 things together well, you're going to get better route adherence, so better rep run times, more reliable ETAs and better asset utilization.

Well, thank you very much for that, and thank you, guys, for your great answers and this great panel. I've learned a lot myself in the last hour about the applying AI and machine learning to this whole issue. It's now time as I promised to bring the audience in. We do have some audience questions already submitted. And if we'll get to as many of those as we can time permitting, but please do continue to submit your questions by clicking on that Q&A icon at the bottom of your screen. So I'm just going to throw these out and see who wants to take them. This question is what's the most practical first step for a fleet that wants to move from static planning to more data-driven service time predictions. I see you nodding Cyndi, so I'm going to throw this one in your direction.

Well, I think the practical for us is to start capturing and using actual service time data that's coming in from the field. A lot of fleets have the data. They can pull it in from telematics, cameras, other [indiscernible] apps, but it's not systematically being used. If you aggregate that data and apply to planning, even in a super simple way, you don't even have to use machine learning, although it's better. You can start to replace assumptions.

Okay. Now the question says, do you need a large amount of historical delivery data to benefit from machine learning? Or can smaller fleets still see [indiscernible] Sergio, why don't you take that one?

Yes. Sure. Thank you, Bob. You don't really need that much amount of data. Even smaller fleets can benefit from this. You can see that the models are going to be improving over time. So look, what we're always talking to our customers. We say, look, let's start securing your routes. And we're going to learn a wealth of information that we can actually utilize to improve your -- to improve basically your route planning. And that's where the value starts. Cyndi mentioned it, measuring the actual service times. You cannot improve what you don't understand. So by starting to capture and using the data that you already have, it's probably the best source of wealth that you can have to actually improve your operations. So let's start there. And even -- like I said, even it doesn't matter what fleet size you have. It is important that you actually are capturing that data to learn from it and use it in your planning operations. .

Clear wants to know what is the most common mistake that fleets make when they're trying to modernize delivery operations with AI. Sergio, you look like you're ready to answer that question.

Sure, sure. I'll jump in. Look, trying to do too much at once. That's the biggest problem and trying to actually boil the ocean. The best approach, in my opinion, is to start with a focused use case. And that once you actually have that case defined, now you move on to prove the value and scale from there. So you can actually start with a small operation in a few vehicles to figure out, okay, this is what I'm going to be doing. I'm going to be learning from these drivers. I'm actually going to start communicating with customers to see what the value of doing such action will actually bring to my business. Is it actually reducing time. Is it actually improving the fleet utilization. Is it actually improving our customer satisfaction. And then once you have actually that proof, then you have a platform to actually scale from there.

This question says, how can drivers get involved with how their data capture is being made and used? I like that question. We have to keep bringing drivers back into the picture. We can't forget them. Cyndi, why don't you take that one?

I think people do forget the drivers was such a critical piece of this, right, involving them in the conversation first and foremost. These people driving the trucks and quite frankly, sometimes they have the best ideas. Again, what looks good on paper may not be practical in execution. But one, explain to them what you're doing; two, ask them what the barriers are to collecting the data; and three, ask me if there's a better way to do it, right? They may say, you know what, pushing buttons this [indiscernible] hard, could you turn on [indiscernible] part. They may say, if you collected this metric or this piece of information, while I was at a customer, I can reduce the amount of tribal knowledge that's needed to be a service customers. So have -- just a and have the conversations with them. We don't do that often enough, and so many great ideas come from drivers.

Okay. A real quick 1 on this one. Maybe I'll give this 1 to Sergio, just very quickly. How will machine learning continue to involve in this context?

Well, it helps tremendously because it helps us actually learn from what actually the driver is doing. So in that context, as Cyndi was explaining, look, if you involve the drivers right from the beginning and you tell them why you're actually collecting data when he's utilizing a mobile device, -- and you're telling him, it is important that why we're actually measuring when you arrive versus when you complete your services is actually to provide you with better routes and that will basically improve your driver satisfaction in a long shot. So -- and at the same time, that data will be basically used and the driver will be able to see how that actually affect is impacting the routes. And if you actually with data, again, with data visibility, a, Mr. Driver, look at this, this is actually the way you're actually securing, this is the way we planned it. It's actually very close. So we're actually doing well, and the driver can come back and say, you know what, now that you're actually coming back with that. I told you I can do this route in 6.5 hours instead of the 8 hours that you gave me. Let's put on more because probably that improves also the driver satisfaction to actually do more work maybe you can incentivize driver to be more efficient [indiscernible]

[indiscernible] I must cut us off at this point. It's such a fascinating discussion. I wish we had another hour to talk, but we don't thank you for that answer, Cyndi, as well. Both of you guys for the -- for your great participation. We do have time for just 1 final question. And here it is. We know the delivery teams are being asked to improve stop density, on time and in full performance, customer visibility without adding cost or complexity. We know that. What is the 1 shift, though, that leaders need to make now to move from reactive execution to more precise data-driven operations and what is the 1 action they should take in the next 90 days. Cyndi, why don't you go first on that one.

So, that was a pretty tough but [indiscernible] right there. Thank you. I think here's the reality. You have to make a shift from kind of these strategic static plans to continuous data-driven operations. You have to use the real auto data to improve decisions not only in real time, but in planning time. And the next 90 days, focus on a single use case, something -- I'm going to say simple, but it's not simple, like ETA accuracy, right? Start turning that execution data into action. If you focus on one, that's how you can get impact fast.

Sergio, what do you think?

Again, as Cyndi mentioned, in your shift is planning from assumptions instead of planning on assumptions, you actually plan on actual delivery behavior. That's the #1 shift that I will mention. If you have something to do, and I'm going to give you an action item on the next 90 days, it's start capturing and using that real delivery data. That's it. capture it. And that's the very first foundation to actually start looking into service time predictability and execution variability. So that's the foundation for improving your planning [indiscernible].

Well, again, guys, this has been a great discussion. Thank you so much, Sergio in Cyndi for this excellent presentation for your answers to my questions as well as those in the audience and thank you, audience, for posing them. This has been wonderful. Thank you so much. We have for you an interactive demo real-time visibility that drives delivery performance. You can access it by taking a quick shot of that QR code. You'll be able to see deliveries in action. Scan that to explore the experience in real time. And we also want to bring to your attention. These Episode 2 in this 4 webinars this year from Descartes under the common theme of AI Exchange. This time, it will be about AI Agents for Fleet Performance Management. The date is June 23, the time is 11 a.m. Eastern. Again, there's a QR code for you to take a shot of if you want more information to save your spot. However, if you don't have time to take that QR code or the 1 before, which has already disappeared, obviously, don't worry about it. All that will be provided to you attendees at the conclusion of this webinar, which, of course, is right now. Thanks again, Cyndi. Thanks again, Sergio. Thank you so much, audience. Everybody, have a great day.

Appreciate it Bob. Thank you.