Intelligent Freight: Protecting Margins with Predictive Pricing | Episode 65

All right. Hi, Dawn. Welcome to the Banyan Tire Tracts Podcast. How are you today?

 

I'm good, April. How are you?

 

I'm doing great. Thank you. Welcome back. This isn't your first time on the show, right?

 

It is not. I've been on at least once before. Yeah, it's been a while though.

 

Awesome. Well, it is the first time here on a podcast, so I'm happy to join you here. We'll just get started with a little bit about you for anybody who might not have seen a previous podcast.

 

Great. Well, I appreciate that. Thanks for having me on again. My name is Dawn Favier. I am currently the president of Triumph, Intelligence, was previously the CEO of Greensscreens.ai. For those of you that are not aware, Triumph acquired Greensscreens in May of 2025 to really form the Intelligence segment of the Triumph business along with ISO, Isometric Technologies who they also acquired in December, just a year ago, I think this week. Together, we have formed the Intelligence segment.

 

Just a little bit about my background. I was at Greensscreens from the beginning as a Chief Product Officer and CEO, eventually just really slid more into the CEO. Prior to that, I spent 20 years working in the TMS space, mostly in product strategy, product management, go-to market strategy, and started my career as a shipper. Worked in the inbound transportation department to major retailers located in the Northeast and migrated over to technology after about nine years of working as a shipper.

 

That's awesome. We talked about your background a little bit. What as a shipper, right? Where have we moved from to, did the shipment deliver, how important was that to you as the shipper? What does that look like these days?

 

Yeah. Well, it's changed a lot. I mean, look, I'm going to date myself here. When I first started moving freight back in the early 90s, we literally had maps with map tags and sticky notes. We were doing inbound freight to our distribution centers. Our vendors, our suppliers would call in. Or we had a routing guide. But once it was over 5,000 pounds, or 750 cube, they would have to call in, and we would manually try to build multi-stop loads, right? That was how we were incentivized as load planners. But we were writing things down on sticky notes and we had sticky notes all over our credenzas and map tax. We've certainly come a long way. I think we implemented our first TMS, I want to say roughly 1994, something like that. We were leading edge on that. But I will say, in the TMS technology that's available today as compared to what was available in the early and mid-90s is light years different.

 

Everything was manual, right? If we wanted to run distance, we had PC Mylar and we would have to do that. All of the tracking trace was manual. Or there was EDI. We'd get YD14s, but it was still a lot of phone calls, calling the carrier, calling the dispatcher. Did you pick up? Where are you? What's your ETA? Yeah, it's amazing how far we've come.

 

Yeah. It's so funny you say that, right? Going back to ’94. But I think even in today's industry, even in the last five to six years, just how the freight industry has evolved. Going from really that historical reporting into predictive AI and predictive intelligence. That's really kicked up within the last couple of years. Can you talk about how that's made a change in the industry in the last couple of just last five to six years?

 

Yeah. Really quite a bit, right? The predictive AI. Of course, that's what we do with Triumph Intelligence and we're really focused around coming transforming freight pricing, right? We're looking at this from a pricing perspective and really transforming that from being more reactive to more predictive, using AI and AI-driven pattern recognition. The goal of that is really to identify trends and anomalies in the data and to serve up some really actionable insights coming from this massive fragmented data set, right, which is helping to improve rate accuracy and decision making for predominantly freight brokers and also shippers today. There's so many other ways that AI, predictive AI is being used.

 

I mean, think about even in the track and trace world, right? Being able to do more. If you get that late ETA message and being able to predict forward, okay, what's the impact of that? Or to be able to identify risk of service failures even before they've happened to use that big data and ML to do that.

 

How do you feel the industries understanding and how do you get people comfortable with that type of data and everything that you're reporting on today?

 

Yeah. Look, I mean, I think this type of predictive AI, machine learning, predictive AI over the past couple of years has become more broadly adopted and more accepted. Certainly, when we started Greenscreens in 2020, it was so novel. Nobody was doing it. Nobody really understood it. A lot of what we had to do is education, right? Why is this better than some of the more historical ways that folks had figured out market rate intelligence, right? Which has generally been more backwards looking in nature, and looking at median rates going back three days, seven days, 10 days, whatever the case may be. Then I think the other hesitation, again, going back five or six years ago was people's hesitance to give their data. Data is so critical to AI and machine learning. That was also something that was really difficult to overcome.

 

Now, there's no shortage of AI applications today. There's a huge proliferation. We’re clearly in an AI bubble. I think what I'm speaking about is machine learning and pattern recognition and things like that, which is very different than agentic AI and LLMs, everything that we're seeking and proliferating the market. I swear, every week I hear about a new AI, agentic AI vendor, right?

 

Yeah. Yeah. How do you then redefine the predictive intelligence through that pattern recognition and stand out? At what point does a TMS being a tracking tool start behaving like a forecasting engine type?

 

Yeah. I mean, look, I think, that we talk about TMS of the past and how it compares to what we have today. I think the TMS of the future really becomes more of an operating platform, or an orchestration platform that is connected to some highly specialized apps, with predictive intelligence being one of those, right? Sure, maybe some TMS vendors might create their own predictive apps, but I think it becomes less of core feature function and it's really more of a bolt on, right?

 

Yeah.

 

That's how we see the TMS of the future. Look, I mean, if you think about taking this huge volume of data, so Triumph, right? Triumph touches about 65% of the truck load freight in North America through our payments network and our factoring business. You're talking about verified audit quality data. The 65% plus with our – the Intelligence, or formerly Greenscreens connections to our customers’ TMSs. So, we're integrated with Banyan and 12 other TMSs. We're not only getting that, after the fact verified audit quality data, but we're also getting every time a load is booked in real time, that is flowing to our engine, right? We're continually retraining and retraining. If you think about – let's talk about pattern recognition. Human brain, pattern recognition allows us to relate what we're seeing in real time to memories we have of what has happened. Essentially, AI and ML are really just trying to replicate that. The pattern recognition is certainly training off of all of that historical data, but it examines new data, compares it to the existing data, or its memories, and learns from the data and can easily identify known patterns.

 

I think more importantly, it's able to recognize new patterns in previously unseen data very, very quickly than a human might. It might take a human hours or days, or even longer to be able to. But also, able to identify patterns that don't conform to known norms, right? Within, again, its memories, which is the data that lives in there. I mean, look, there's a very stepwise process in how to do AI and ML and the first step is gathering the data. For us, it's that load data, whether it's the booking data, or after the fact that the paid transactions with API and SFTP and any number of ways of connecting that data.

 

Then there's pre-processing, and it goes back to garbage in, garbage out, right? To have a really good algorithm, the ML algorithm, and you really need to cleanse and eliminate the noise and the outliers. We have a process that goes through that. When we onboard a new customer, we often flag about 35% of their data as potentially dirty.

 

Wow.

 

Then we'll work with the customer to resolve that, to minimize it. But the key is protecting the algorithm from noise and outliers. Then once we've done that, then it goes into the analysis phase. Through the algorithm, searching for informative features and patterns. Features, if you think about a load, all of these different attributes are called a feature, right? Some of them are explicit mode, equipment type, origin and destination, right? Those are all explicit, but then there's also some derived features. Transit time, day of week, lead time, all of these things that can be derived from other features that are also considered as the algorithm’s going through it.

 

Then it goes through clustering and classifying. It looks at this massive mountain of data, and it will start clustering and classifying data based on those individual features. We look at several dozen. I think four, or five dozen different features of the load that we will look at as we're doing that. Then finally, it's understanding, okay, I've classified, I've identified these patterns. What are the patterns telling us about the problem that we're trying to solve, which in our case is for cost, and then making that prediction based on all of that iteration?

 

That is a lot of great information. Something that you said earlier made me think of. I always joke with my clients, especially now in the AI world and these agents and things. Anytime I have a tickle in my brain, I'm like, what could that be reminding me of in this scenario that I'm thinking of? But those are some of those early signals, that AI is going to catch that way faster, or the predictive learning than I am ever going to remember, and I might know a month from now, or if I see it again, oh, that's why that was tickling the back of my brain type thing. Yeah, it's going to learn it much faster than we ever do, right?

 

I think you talked about understanding the anatomy and some of those early signals. But what are some of those real triggers, or the micro signals that might be jumping out to that intelligence that we might miss as a person?

 

Yeah. I mean, look, I think there's some – Well, look, one of my favorite anecdotes is we had a customer that they were still in their implementation phase, but they pretty much owned a lane out of somewhere in suburban Iowa. A manufacturing facility somewhere out there. They knew the lane. They ran it all the time. They ran a rate through our system and the rate came back about $500 more than they knew the rate to be from running the business. As they should, they called us and they said, “We got an issue.” They opened the ticket with us and we said, okay, we'll look into it. Really, before we even have the chance to dive into it, they called us and they said, “Never mind. The shipper doubled production at that facility this week. They didn't tell us.” All of the capacity that would normally service that market has been absorbed, or consumed. Now we're bringing capacity in from other markets. That $500 more expensive rate was actually accurate, because we were deadheading equipment in from another market to serve that market. The algorithm was actually able to pick up on it before the humans even knew –

 

That's amazing.

 

- what was happening. It was because of that real time nature of the data, right? It can see, oh, suddenly, there's a lot more volume moving here, which is going to impact supply and demand balance, which is going to drive the rates off. Look, I think humans, I think somebody who has been doing the job of a freight broker for 10 years, 15 years, 20 years, they have a lot of track with knowledge, right? They may have a lot of this stuff stored in their head. They know that a load that's picking up on a Friday and go in 700 miles, but can't deliver until a Tuesday, they're going to pay a lot for that, right? Or a 400-mile movement picking up on a Friday, they can't deliver till Monday. They know some of these and things instinctively, but those are, think about a new rep who hasn't been at the job for 10 or 15 years, doesn't have that tribal knowledge. What are they going to do? They're going to go to a more experienced person.

 

The machine learning doesn't need to do that, right? The machine learning will find those patterns in the data and be able to predict it. Again, where we've seen to be really powerful is in unseen data. If you are somebody who has to move freight on a lane, you've never moved before, right? You don't have carrier history there. You don't really know the markets. We're able to also use, we call it a similarity model, where we can look at both adjacent and non-adjacent lanes, or markets that behave similarly in order to be able to always predict a rate, even if it's on unseen data. Something we've never seen before.

 

That's amazing. I think about some clients that I've worked with in the past who are seeing high-cost area charges that they hadn't seen before, especially around seasonal changes, even protecting from freeze. Calls out a lot of assessorial type data that we might see in the day to day on the frontend side of things, excuse me.

 

Yeah. Well, so today with our product, we are really only predicting line, haul and fuel, right? Because there is so much variability in assessorial charges. We're really only predicting line, haul and fuel. However, there are characteristics about certain shippers. We will predict to the five digits of code. If historically the data tells us that there is a shipper located in that five digits of code that has bad behaviors, or the product that they're shipping, the commodities that they're shipping are difficult to handle, or has not, or needs special licenses and things like that. A lot of that stuff gets inferred from the data itself, again, based on what is the specific zip code? Who is the shipper? What commodities are being moved, right? That type of thing.

 

Yeah. It sounds like, that type of behavior helps sharpen a user's ability to quote accurately and confidently, if you –

 

Yes.

 

Yup, yup. Beautiful.

 

Yeah, absolutely. That is our mantra. Transact “with confidence and transact confidently.” Yes.

 

Awesome. What would you say, what are some of the biggest misconceptions that a shipper in 3PL might still have about predictive pricing?

 

Yeah. I think, look, nothing is 100% perfect, right? We do measure the accuracy. Think about it like playing darts, right? When you're playing darts, you're not going to hit the bullseye every single time, right? If you're pretty good at it, you're going to hit the rings around outside the bullseye, but it's very rare that you hit the bullseye every single time. It's the same thing with machine learning and predictive pricing, but we do measure in real time through the application, very transparent with our customers about the accuracy of our prediction. When we predict a rate that we can then go back and compare that to what did that load actually then move at, right? What did it get booked at and measure the accuracy of that.

 

Right now, we're in the single digit margin of error. On average, across all of our customers, roughly 6%, 6.5% margin of error. Some of our larger customers are 2%, 3% margin of error, right? But we don't bring a customer live, unless our margin of error on prediction is less than 10% always. That's one misconception is it's not perfect. Okay, well, it's never going to be perfect. I think the other misconception is that more data is better. That's not necessarily true. Now, mind you, we do, as I said, touch 65% of all truckload rate. It's about 74-ish billion dollars of data that we sit on. It's less about the size of the data. It's more of a vanity metric. What are you doing with the data and what is the technology that you're applying to it, i.e. machine learning and pattern recognition and the way that you're training your models and how you're protecting the models from being something that's called an overfit for the problem it's trying to solve.

 

Okay. How does predictive pricing, how does it help a team avoid margin erosion and underpricing, especially in unproductive, unpredictable markets?

 

Yeah. Look, I mean, think about this. If you are quoting a customer and you have a margin target of, let's just for easy math, say 10%. Hopefully, your margins are more than that, but let's say your target is 10%. You're quoting the customer without a truck in hand, so you don't yet know what you're going to pay. But through your market intelligence, you're presuming you're going to pay $1,000 for that truck. Then you quote your customer $1,100, because you have a 10% margin. But what if your presumption of $1,000 that you're going to pay is wrong by 20%, or 15%, and you actually end up paying $1,200 for that truck, but you've already quoted to the customer $1,100 and you've won that business on $1,100. You're either going to lose money, or your customer is going to be really pissed off.

 

When you go back to them and say, “Hey, I can't do it for $1,100. I need to charge you $1,300,” or whatever the case may be. We've actually been able to really tie that margin of error to margin, and what we've shown is for every 1.5% improvement in our margin of error can lead to a 1% improvement in margin, right?

 

The other thing that we've been able to show is we did a study of 730-ish users of our product based on the load data. We know who the person is that booked the load and we also know what users are actually using our product and what users are not. What we were able to show is that the users that were using our product were seeing 13.7% improvement in margin over those who are not. I mean, that's the percentage improvement, right? The difference was about 2.4% margin, but it was a 13%, 13.7% improvement.

 

Yeah, absolutely, absolutely. It sounds like these users are using multiple data streams and even predictive intelligence is going to do the same thing, right? It's going to get more accurate, or stronger when you blend those multiple data streams. How does cross-referencing shipment behavior, market pricing, assessorial trends, tracking signals, how does that create a more reliable forecast for a user?

 

Yeah. I mean, again, if the closer you can get to reality, the more you're going to hit, your margin realization is going to be better. Your spend analytics are going to be better, because you're going to have a better idea of what you're going to pay. Even now, I mean, look, since about Thanksgiving, we have started, you've probably heard about it on the news, tender rejections are going up and rates, at least in certain regions are starting to go up and we're watching how our algorithm is adapting to that. Were we under-predicting for a little while? Yes, we were, right? Because it did take a little bit of time for the algorithm to adjust. But even though we were under-predicting, it was only by 1% or so. The market was shifting a lot more than that.

 

It just helps you get a lot more predictability into your business. Rather than really being caught behind on some of the market volatility and the swings, or at the very least, to be able to react to them more quickly when they do happen.

 

Okay. Yeah. That sounds like a really recent moment where those dots are exposing a problem, essentially, connecting the dots in the learning, especially this time of year where prices are going up retail and things are going to change. Are there any other moments where opportunities are found in scenarios like that?

 

Yeah. I mean, I think a lot of opportunities are found in scenarios like that. Look, if I go back to the history of the company, we started in 2020 and really brought the product to market in 21 when freight rates were astronomically high, historically high, and margins were historically high, and it was a great time for everybody. At that point, it was really all about minimizing the volatility, maximizing, or optimizing the buy rate, what you're going to buy the freight at. But then, the market shifted. We've now been in the longest freight recession that we've ever seen. But yet, we didn't really have to change our product significantly. It just became the margin piece of it became more important. But being able to tie that precision of the predicted buy rate to the margin, and it became about margin preservation more so than optimizing that buy, because the rates were so low, right? How do I improve my win rate and how do I improve my margin and my volume and all of that?

 

Right. Which is a great lead up to my next question. We talked about a little bit earlier, well, you've talked about garbage in, garbage out. We hear that a lot in the industry. How critical then is data cleanliness and normalization and making sure that's clean when you're merging operational data with pricing and behavior signals?

 

Yeah. It's incredibly important. I mean, if you think about the fact is take one lane and you've got – I talked about we're getting rid of the noise and the outliers. But if you've got a hundred loads that move on a lane, but the price spread between them is $1,000 and $1,400, that's a pretty big range, right? If you were to look at that on a scatter plot, maybe that $1,000 and that $1,400 is their outliers, right? That's the data that needs to be cleaned. We also find things like, just user error. One of the filters that we apply is anything under 5,000 pounds, because we're focusing on truckload, right? Anything under 5,000 pounds will automatically flag as potentially dirty, assuming that it's either an LTL, or a partial that got misclassified as a truckload. Those will be flagged and excluded from the algorithm, because the thought being that whatever price is associated to them, that doesn't represent a valid truckload price.

 

However, we do know, working with some of our customers, that a full truckload of empty plastic water bottles weighs about 3,500 pounds. So then, we're able to hone in on that and adjust those to say, okay, for this shipper, or from this location, 3,500 pounds is valid. We might lower that threshold to 2,000, or whatever the case may be. Or looking at distances, right? If the distance, if we're seeing with a multi-stop, if we see a lot of circuitous miles, or something like that, those are examples of things. Or sometimes the 6.00 Friday night, a carrier fell off a load, it's a really important shipper and you've got to move the load at any cost, right? You're going to pay a premium. Those are the types of things that we're looking to identify and isolate and protect the algorithm from.

 

How do you design those systems and how do those alerts get delivered to a user without overwhelming them, or creating any type of fatigue?

 

Yeah. The alerts themselves don't actually go to the end user. It's just the system is self-correcting. It will flag that data market as invalid, but not throw it away and hold on to it. It does have the opportunity to self-correct. Particularly with the outliers from a price perspective, because it has its own separate algorithm that is based on standard deviation. If we throw something out as an outlier, because it's $1,400, which is beyond the standard deviation, but we suddenly start to see a trend of it’s getting closer to that $1,400, that's no longer an outlier, so that would be brought back in.

 

Part of the goal, April, is to not overwhelm the user with – and these are things that a user would likely have to try to figure out by themselves looking at the data. We don't want to overwhelm the user with these alerts. Now, I will tell you, internally, me, our customer success team and our pricing analysts will be alerted if they see – if the system sees a major increase in the amount of data for a customer that's getting flagged as invalid. Then our pricing analyst team would be alerted to that, and we'll be able to go in and look and say, okay, what are the flags that are increasing, that weren't there a week ago? Then we can have a conversation with a customer to say, “Hey, what's going on in your business? We just noticed this with the data.” It's not a one and done thing. It's really a very iterative thing, with keeping that data clean.

 

I always like to say to customers that data cleanliness really starts with your standard operating procedures, right? You shouldn't have standard operating procedures designed around data cleanliness. Sadly, most TMSs aren't necessarily designed with data integrity in mind. They're designed with reducing clicks and keystrokes, which allows for a lot of often, allows for a lot of dirty data to get in the system. How do you establish the training and the standard operating procedures that's going to promote cleaner, better data coming into the system from the onset? When that can't happen, we do have the means to make sure that we are cleaning it and protecting the algorithms from the dirty data.

 

Yeah, that's a great call out. Yeah, it really begins at the front end. The best data that you can put in is how that's going to learn. Talking about AI models and maturing, how do you see them blending operational intelligence with pricing intelligence?

 

Yeah, and look, we're already doing that today. We talked about some of the agentic AI, or some of the other tools that are email, chat, email bots, email agents that will sit in your inbox, where either the agentic AI is interpreting the emails, or the communication that's coming in and interrogating that data and then calling to our AI, which is taking, consuming that request, if you will, generating the AI, the predictive price, and then feeding that back to the agent. We have a lot of customers and partners that we're doing that with that today. Quoting customers, booking carriers, negotiating rates with carriers and customers. Yeah.

 

Now, we're actually even also bringing performances. A broker can see how they perform relative to their peer, as well as how well their carriers are performing on time pickup, on time delivery, bad bounces, all of those types of things, right? Also being able to factor that in as we are interacting with the other agentic AI and workflow tools that are out there.

 

Awesome. Awesome. I'd like to talk next about some real-world applications of pattern recognition. Detention and assessorials are predictable once you spot the patterns, we talked a lot about pattern and predictive learning. What does early detection look like in practice?

 

Oh, that's a pretty deep question. What does early detection look like in practice? Well, again, I'll go back to the story that I told about that manufacturing facility that doubled production and we were sucking up all the capacity, right? The sooner that you can find a new trend, and that's the job of pattern recognition is, is there a pattern that doesn't conform to the norms, and identifying and throwing up a flag on that. The sooner you're able to recognize those abnormal patterns, the sooner you're able to adjust and not be, again, overpaying, underpaying, losing money, all of those types of things.

 

Talking about some of those signals in the data, so fraud risk, suspicious behavior, how often are those types of things hiding in plain sight and how are they identified?

 

Oh, yeah. Absolutely. They are offering and there are some great tools out there around fraud protection that, yeah, able to catch it before a human can even identify it. It's pretty incredible, some of the fraud protection tools that are out there.

 

That's awesome. That's awesome. How can predictive intelligence help teams “with higher confidence?” Avoid rebrokering, and protect those margins that you talked about earlier.

 

Yeah. I think it really comes down to a lot of the things that I said. Knowing specifically where you're going to buy, right? All of our customers see two things. They see their verified buy rate, which is a machine learning model that is trained specifically for them. They also see the market rate, which is the aggregated anonymized data of everybody in the Triumph network. Because not all brokers have the same buying power. The market rate might be $1,000, but broker A may have never moved that lane before, may have no carrier relationships there. Maybe they're going to pay $1,200. That is one of the examples of you're not only going to see what the market is doing, but how your buying power compares to that.

 

Some of the other things that we can do, as I mentioned, you can see how you perform relative to your peers on performance metrics, so that you know how conservatively, or aggressively should you bid, right? If you see that you perform in the 99 percentile of all of your peers on a particular lane, maybe you can be a little bit more aggressive in bidding that lane. Related to performance as well, we can show you not only capacity, or carriers within your network, but also capacity, or carriers outside of your network and how well they perform through the data as well. Protecting you against service failure, as well as earlier on with the price, protecting you against margin compression as well.

 

You talked earlier about scenarios where a user might be shipping something that they're reacting to a little bit more, right? A hotshot load, getting something out quickly, pricing being higher. How does predictive intelligence turn reactive freight work into a proactive one?

 

Yeah. Again, I think it's just the real-time nature of the data that we're getting. It's the real-time bookings. It is the real-time payments. We know right at this moment, or today how freight is actually moving, so you're becoming more proactive with that. Just to have a better understanding of what is the market? If I need to move this today and be very reactive to it, what is the market? Because what the market was doing last week is not helpful, when you need to move something today, or at the last moment.

 

Some of the other things, and again, I go back to the less experienced rep, we have negotiation tips that are built into our application, too. As somebody is on the phone negotiating with a carrier, or with a customer to say, “Sure, I can help you out, but it's the last-minute shipment and I'm going to pay a premium for this.” Or negotiating with a carrier. “Look, man. I know this is last minute, but you're asking me for double what I've ever paid. The most I've ever paid to move this is X and you're asking me for two times that.” Just arming them with the decision tools to help negotiate and protect them and be more proactive.

 

What does the future look like for predictive freight intelligence? What's next?

 

Yeah, I think we're going to see more and more of this. As I said, we've just recently started bringing in performance as well as cost, right? Starting to understand the price versus performance tradeoff, or cost versus performance tradeoff. Do you really pay a premium for better service, or not? Starting to look at that more holistically. Really looking to connect the entire value chain. Shippers, brokers, carriers with more of this predictive freight pricing, to make the way that the industry buys and sells freight truly more dynamic and predictive than what we've been able to do in the past, because everything has really been more backwards looking reactive to the markets, instead of being able to look forward.

 

Ideally, I think we have a vision of, hey, wouldn't it be great if we could move towards index linked contracts, so that the transactional negotiation between a broker and a shipper doesn't need to exist. Where everybody can align to an index that is highly accurate, but predictive in nature and really adaptive to what the market is doing, and index off of that on a cost-class basis. But yeah, really trying to remake the way the industry buys and sells freight through machine learning and AI.

 

What do you think is going to be the biggest barrier to that adoption in the industry? Trust, data sharing, operational changes, lack of awareness? All of the above?

 

All of the above. It's funny you ask that question, because usually I get to moderate, or speak on panels often. One of the questions that I almost always ask when it's on this topic is what organizational changes do you think you need to make in order to fully adopt dynamic pricing in AI? It is a lot of those things, right? It's awareness, education, it's breaking bad habits, or old habits, sorry, old habits. It's making sure that your people's incentives, their incentives are aligned with what your strategy is, right? Making sure that you have strong change management to drive the adoption and ongoing education and also, it's all of those things.

 

I think the reluctance to share data is diminishing. I think more and more companies are realizing that we’re better together with the data. As I said earlier, more data doesn't necessarily mean better outcomes. It just means more data. I do think when you have more data, or a large volume of data and you are applying technology, like machine learning, it is a better outcome. I think people are starting to realize, even some of the very largest group brokers that may have 3 billion dollars or more of their own data, but they only know what they know. If they want to start doing business in places that they've never moved before, they don't have access to that data. Again, the pattern recognition and the similarity models of the ML engine helps to overcome that as well.

 

That's fantastic. That's awesome. All good information. It makes me think back to last week. We had a company event where we're talking about AI is not the future. It's here. It's now. We're all adapting to it. Widespread here, adopted already at Banyan. I just have a few more questions. What is one piece of advice that you would give to a shipper, or 3PL who is still relying on historical averages, instead of predictive behavior?

 

Yeah. I mean, look, I think start small would be – integrate predictive tools into some of your existing workflows, but start small, right? Be very focused at first. As an example, we had one customer that they were doing automated quoting using an incumbent tool. They were winning absolutely 0% of what they bid on, or they were losing 100% of what they bid on. The approach that they took is they said, “Okay, we’re going to keep this isolated to the freight that I'm auto quoting to a subset of my shippers. We're going to start there and prove success there.” I think that is key to adoption is involving your champions, or have some proving success, or early wins, involve your champions early, the people who are your power users, the people that folks look up to that are going to be the loud voice to talk about the success of the project. You need to have strong change management. See, I'm some tumbling on my words.

 

That’s all right.

 

Grand stage management. Or no, strong change management, because people are just going to change their habits, because you told them to. Strong change management, it's going to be necessary with any technology project, but certainly with this one.

 

Yeah, you definitely started to answer my next question right on. If someone wants to take their first step towards that, whether operational or in pricing, where should they start and change management is number one, right? Just getting everybody onboard. But any other little tidbits, or starting points that you can recommend?

 

Yeah. I mean, look, know what your goal is, right? Know what your desired outcome is. Too often, I've seen people implement new technologies just for the sake of implementing new technology, without really knowing what the desired outcome is, or what value it's going to provide. Then months down the road saying, well, this was a failure. But when you say why, what we're hoping to achieve, they can't really answer that question. I think it's change management, being clear on your goals, aligning your goals to your employees and all of those things that I've already said, like alignment, and just be very clear about what the vision is.

 

Yeah. It's more than just going out and downloading every AI app, or whatever you see on the market right now. It's really understanding exactly what you're trying to accomplish with those. I think that's great. That's awesome call out.

 

Exactly.

 

Yup. I have one more question, essentially, for you. We talked about a lot of the predictive intelligence and where it's going and what you're thinking, but where do you see that evolving over the next three to five years? We've already seen how it's evolved so quickly in the last even year alone. What does that look, do you think?

 

Well, on the predictive AI, I think we're just going to continue to see that grow, right? Solving bigger problems, or more integrated problems, if you will. As I said, our vision of connecting the ecosystem of the shipper, carrier, or broker, each of them has a slightly different pricing problem, but they're all interconnected and interrelated. I think we will see that begin happening for sure. I think the AI, the ML models are just going to continue to get stronger, particularly as we get more data consolidation.

 

I think AI in general, we are definitely in an AI bubble, particularly with the agentic AI. I think the bubble is going to have to burst at some point, right? I think there will be some consolidation. But there are a number of really powerful agentic AI tools out there that are really specialized and do a great job at very specific things. It goes back to my comment earlier about the TMS of the future is more of an orchestration layer with a bunch of bolt-on specialty applications. I think agentic AI is going to be no exception to that. But I think there needs to be optionality. I think there will be consolidation, there has to be. But technology and particularly this type of technology is not one size fits all. I think there has to be optionality available in the market.

 

That's awesome. That's great. I'm excited to see how everything grows over the next couple of years, right? We've already seen a lot of it. I think that's going to be it for all of my questions today. Anything that we didn't touch based on that you'd like to bring up during this call?

 

No. I appreciate having me on. If anybody would like to learn more about Triumph, Intelligence, you can find us in on LinkedIn, Triumph Market Insights, or Triumph for Brokers, or on our website, triumph.io.

 

Awesome. All right. Well, thank you so much for your time today, Dawn. We'll talk to you soon.

 

Thank you. All right, talk soon. Thank you.

 

Thanks. Bye.

 

Bye.