C.H. Robinson Worldwide, Inc. (CHRW) Earnings Call Transcript & Summary

[Audio Gap] at Citi, he's going to be asking the smart tech questions. And then we're thrilled to welcome C.H. Robinson to our conference. We have Dave Bosman, CEO; Damon Ling, CFO; and Arun Rajan, Chief Strategy and innovation Officer; informally CTO, I suppose is fair. It's interesting. So just before we started, Dave and I were talking about the idea of a transportation and logistics company at the tech conference might be a bit strange to some folks. And yet those who are familiar with the Sage Robinson story certainly would not be surprised by it.

Just looking back to the last couple of years, if we think about your North American Surface Transportation segment, we've seen gross profit per employee up more than 60% at a double-digit rate. there's really been some remarkable things that you guys have been able to do with technology and leveraging AI and specifically, you refer to lean AI. Maybe you could discuss that just kind of as an introduction, what are the changes that you've introduced that have been able to drive those gains? And specifically, it would help, I think, if you gave some specific examples around what have been the challenges that have kind of -- you've had to overcome to get to that?

No. Yes. Thanks, Ari, and happy to be here at your conference. Thanks for having us. For us, as we talked before, it's not strange to us being here because we know the story and the transformation we've done at Robinson over the last 3 years. I'll just kind of frame it and then have Damon and Arun kind of get into it because it is a story. You talked about Lean AI. What we've done is introduced a transformation of a 120-year company -- we essentially make it a disruptor within this industry. And we've done it in 2 kind of vectors. One, a very different lean operating model based on lean principles. This is about continuous improvement problem solving, pace, speed, all things that have been around, but we've introduced into this company, and I think also have been pivoting this industry that has really opened up and allowed our technology, which we're builders, not buyers and always have been. It allowed our technology and our data set, which is pretty proprietary in the largest in the industry. to really kind of collide to make that data intelligent. And then between our lean AI, our -- the best technology, we think, in the industry -- and then our people, of course, which are some of the best largest titans in the world. Those 3 have been Symbionic in the system to at Robinson and has really allowed some of those gains that you just talked about that's what's counter driven at. But let's get into the technology in particular and also our operating culture, which drives bottom line results. But we can do that, Damon or Arun?

Arun, why don't you talk about the tech and I'll wrap it all up.

Yes. So from a technology perspective, like you noted, we've improved our productivity by 60% over the past 4 years. and sort of this journey of we work backwards from what would a tech company do to disrupt this industry, right? So I mean when you work backwards from that, you say, well, a typical tech company would say, well, how do we decouple head count growth from volume growth, right, which is like to make this a very scalable business, right? So some of us came from Amazon, like that's the model that you would do. So the first thing you do then it's like, okay, well, how do you engineer the system to ensure that it's scalable, effectively decoupling head count growth and volume growth. We've been on that journey for multiple years now. And you see the 60% number One thing that's changed materially over the last couple of years is we use traditional software engineering approaches for the -- I've been at the company for 5 years, but the first 2 or 3 years of more traditional software engineering approaches. But over the last couple of years, in the past, a lot of projects wouldn't sort of like make the ROI bar because there's a lot of software engineering effort. This industry is kind of plagued by nuances that are specific to customers that are specific to carriers because we sit in the marketplace, right? We sit in the middle between capacity and customers and unlike a peer-to-peer marketplace, this is B2B marketplace, which has lots of nuances. And we sit in the middle of that, and a lot of these projects didn't make the bar 2 or 3 years ago. But with AI, and now all of a sudden, entering is massively better. We have 500 engineers, but those engineers punch like their 2,000 or 3,000 engineers as they start to use coating tools. That's on 1 side. The other side of it is we've engineered a platform and a genetic AI platform that basically allows us to capture the collective knowledge of the company, the context of the company into our Agent platforms. Now all of a sudden, you're not building software in the old way, right? We're not engineering every single rule or every single SOP into software, it's in a context layer, right? So in a way, we're -- the company is being programmed in English, right? I mean it's probably the simplest way to think about it with the advent of -- so this agent harness that we built on top of the LLM has created this massive acceleration in our productivity in the last couple of years, combined with a lean operating model. So that's 1 side of it. So that's the productivity side of it. And the other side of it is as a marketplace, what do you do, right? There's pricing on 1 side, how you price to customers and then the on the costing and the capacity side, it's how you procure capacity efficiently. So there's a in the marketplace, it's kind of like you make the spread. So this notion of personalized pricing that drives our pricing algorithms. So we have a machine learning-based algorithm at pricing, which would be had for 10 years. But as the data compounds and grows, this is traditional AI, right, I'll call it, classical machine learning that effectively drives our gross margins by driving the right pricing for the right customer for the right amount of value we deliver. And likewise, on the capacity side, it's about cost discovery and how we how we procure the capacity at the lowest price, the lowest cost to us to put on a particular load, right? So again, if I can just step back and sort of summarize what I said, which is, well, if we were trying to disrupt C.H. Robinson as a tech company, what would you do? You would create this scalable model to drive down the unit economics. And then you would drive this intelligent pricing and costing discovery to drive the best sort of gross margin number. And that then compounds ultimately what do you do with the lower cost to serve like we've accomplished, you parlay that into growth by driving prices lower, right? So that's sort of the formula for driving our business, lower cost to serve, higher gross margins, but you partly some of it back into growth by lowering prices to customers.

And just to put a bow on what Dave and Arun said. So since the -- we've not only benefited with 60% productivity since the end of the year, but we've also seen demonstrable revenue growth. We've seen demonstrable gross margin expansion and then certainly, the operating margin expansion we've realized has been on the back of the productivity. But for us, AI, lean AI specifically, it's more than just productivity, right? We see revenue growth. We see revenue management capabilities, so think better price better procurement of freight is productivity. So we see the benefits of our Lean AI strategy up and down the entire P&L. We know what we're doing is very defensible, right? So we have 450 engineers that builds our own tech. This is custom tech for our company-specific problems, right? Very difficult to be able to go off the shelf and try to replicate what we've done. We've estimated that you would have to partner with maybe 15 to 20 individual AI platform companies to replicate what we've done. And even once you did that, you're getting a generic solution set for a company specific set of problems and opportunities versus our customized solutions. And I think the most -- maybe the most important thing outside of that is the cost. Our marginal cost of ownership once we build an agent is close to 0. Whereas if you're using somebody else's third-party tech, you're going to pay by the drink every single time you use their technology. So we've generated hundreds of millions of dollars of operating income value since the end '22 and on an annualized basis, we spend less than $1.2 million on tokens, right. So just so that in an ecosystem, an AI where you cannot find a company that's generated a positive ROI from their investment in AI, we've generated hundreds of millions of dollars of value since the end of '22 with a very immaterial amount of token investment on an annual basis.

And I think for this audience, from an investment perspective, this has been a structural change. Everything that Arun and Damon just said, we are very, very purposeful about where we put the technology, and that's embedded it into workflows. And for us, it was the order to cash workflow. That lends itself very much coming from machine learning into generative AI, and now we're actually doing Agentic within our other business that will come all around. That has allowed us to automate essentially our back end or operational type of roles that it doesn't matter if the market takes off or if the market stays lower for longer, this system is now going to be structural in that change. We won't add in a number of humans in this kind of order to cash type of process. And so we always talk about transactional quotes. We have a mature agent that is doing our transactional quotes to come in with e-mail, used to only get to 60%. Now we do 100% quoting. We do it in 31 seconds. It used to take 17 to 20 minutes. We do it back in a conversational manner. And the point on all of that, that's allowed us to win more freight, see more freight. But it also says that if we're doing 600,000 of those quotes, you can add a 0, we'll do $6 million, and we won't be adding humans to that because that's a mature agent that is placed within that realm and that workflow. So I know that was a long answer, but it covered a lot to show why this is symbionic and why this is structural from an investment perspective, and we feel good about that story.

Yes. No, that's a great overview and a really impressive productivity growth over a short period of time. Maybe diving in a little bit more on the technology stack. I mean it sounds like a lot of this is homegrown, right, purpose-built for Robinson. But how do you kind of see the -- who's as investors here are looking at technology companies like who's gaining wallet share? Is it mostly the infrastructure providers, who's losing wallet share if there are technology vendors from that perspective? And then 1 of the interesting things you said was just $1.2 million of annualized token costs. I'm sure some other companies in the valley would like to be that low in terms of token costs, what we hear. What are you using? Is it the Frontier? Or do you go more the open source route would love for you to hit on that as well?

Yes. Yes. No, great question. So I would say like strategic partnerships or we're a builder culture, right, which means we build our software, which means we will evolve to be an AI and native logistics company, right? So then the question is like, what are the underlying infrastructure providers we use. We've long had a partnership with Microsoft and they're a cloud platform, and they're also the access to multiple models, right, to Azure, we can access multiple LLMs. We have a partnership with Snowflake. Now again, I think that these are infrastructure players, the real sort of value that's coming from AI is from our custom-built harness that sits on top of the LLM, right? So think of it as we have the ability to route a given workload to any LLM, right? We can use -- they can use a frontier model, if it's a complex reasoning. But if it's a simple transactional thing, we can use open source or we can use an older version of the model to a router that will route the work based on the sort of complexity of the work to the appropriate LLM,and in some cases, we host our open source LLMs. So the bottom line is, in the end, because we're customer built shop, and we have our own harness, and we build these genetic workflows, the way we architect our platform kind of accrues most of the value to us, right? Certainly, the token -- there's some token costs that open the AI probably gets a giant share that $1.2 million token spend that Damon described, but there's a bunch of volume into open source. And over time, we will continue to route the simple workloads to open source models, which is why the costs are so low. So then the question is like, well, how do you do it? Just like I think of this as you kind of roll back 15, 20 years to the cloud and when the cloud first came out, I think you had this tendency of people to sort of like say, "Oh, the cloud is here, like I don't have to provision hardware in my data center. So I can just spin up this instance in the cloud and like spending went crazy, right? I think the same thing is happening with LLM where it's like, well, I could easily access this intelligence. So let me point my application at it and just Hey, look, it's great, but it costs a lot. Because of the way we built our harness the way and the way we've engineered our agents, each agent has a very specific purpose and it has a very specific context, which means it has a limited context when context, so it doesn't collucinate. But equally, it's token consumption is very limited. And also we can use an older version of the model or an open source model to do the work. So I think this all goes back to sort of to me that this looks. This era looks like you have to get your platform engineering and your infrastructure engineering right to be able to get the true value of LLM and if you don't harness, we don't create this harness, I think putting end up a lot of labs, rich, right?

Yes. I think, again, for this room, why is it Damon used the word earlier? And why is this defensible? It's defensible because 1 thing running on the back of all of that -- we said this a lot already in the past, is our data set. And we have the largest data set in the industry. It's 100 trillion data points that have been accumulated over time. And that data set is proprietary to Robinson. And when you build a bespoke platform like we're doing with that data set, that makes that very powerful and very hard to replicate even if you're an AI native company to a arteries to do that because you don't have access to that data. You have access to some data, which can be averages of averages but not the level of data that we have and what Erin just laid out is super, super important on why this is a deeper, wider moat.

And there is no hobby spend. on AI at C.H. Robinson right. So every dollar we spend on tokens, every dollar we spend on engineering capacity has already made high probability ROI assigned to it, right? So we don't just give employees or 10,000 employees, co-pilot license and say, "Go, go try to do something creative, right?" The only dollars we spend are based on a very high probability outcome from an ROI perspective, which is why we've had the success we've had.

Yes. And I guess as you think about -- you talked about the productivity gains, which is 1 side of it. But then those incremental revenue opportunities, right, like building a more sophisticated pricing engine, like how much would you sort of attribute these AI investments to driving productivity versus driving revenue? And then maybe just give a sense for the audience, like what are your biggest areas ahead as we look forward in the next few years that AI.

I'll will talk. I'll start and then hand it to Arun. I would say because we let the highest ROI project dictate investment Difficult to say what percentages of our benefits come from revenue management versus productivity versus growth because they're all in the same funnel competing for the same investment dollars I would tell you though the revenue management unlock has been substantial, right? I would say as little as 4 years ago in this industry, just the way in which the industry priced was very unsophisticated, very low frequency, meaning good set a pricing strategy at the beginning of the month and maybe determined at the end of the month that you win or not from that pricing strategy, whereas today, with our technology, with our disciplined approach from lean with the data set that Dave referenced, we're setting pricing strategies on a second and minute basis. So we'll example we use a lot is we come into Monday morning, set a pricing strategy at 8:00. If that's not yielding the volume margin expectations that we anticipated, we could change that strategy tens of times an hour, hundreds of times a day, thousands of times over a quarter, where as little as 4 years ago, you may have only changed that pricing strategy once or twice in a 90-day period of time. just the frequency in which we're interrogating the market from a revenue management perspective and just the ability now to access that data set that Dave mentioned is over 100 trillion data points. Before the advent of AI, the ability to analyze that data was extremely limited. Now with our advanced machine learning with our predictive analytics with generative and agenetic AI. Now we can use a significant portion of that data to drive arbitrage opportunities in our marketplace. And we believe, as I mentioned before, what we're doing on revenue management in the logistics industry, we believe is unmatched.