Episode Transcript
[00:00:01] Speaker A: And I'm going to do the introduction like before this, so then I'll bring you in. Okay. Three, two.
Eric, hello and welcome to the Lodestar podcast. It's great to have you here.
[00:00:10] Speaker B: Hi Charlotte, great to be here.
[00:00:12] Speaker A: So to start the episode before we get under the surface of agentic AI in supply chains, I'd love it if you could tell us a little bit about your background, why you started Zalber and what the company does in a nutshell.
[00:00:23] Speaker B: Sure.
Quick background about myself I have a computer science background. I'm from the Baltic Sea coast. My father was a sailor on hyperkaloid ships which I never thought would be relevant again in my career.
I moved on to study computer science and was sort of in the startup world for quite a few years. In 2015 I ended up being a co founder at Forto, which is a next generation or tech enabled thread for water where the idea was, I think it's known in the industry to say okay, we will use technology to significantly digitalize end to end processes and threat rewarding. And from end to end digitalization automation, the hypothesis was, would come a significant increase in the ability to serve customers well. Great insights, great data quality which would drive retention, higher productivity of course, but also be able to do better revenue management, be on top of every file, every demerge, attention case, charge every waiting time, et cetera.
I did that for almost 10 years as a CPO, CTO. So in the sort of the digital function, building out our own transport management system, building out our own rate management systems, allocation management, customer front ends, APIs, et cetera and it was a fantastic time. I really loved it. The company grew to almost 1,000 people, 20 offices in Asia and Europe. We moved mostly European controlled Seafred and F freight.
It was a really amazing journey. We made lots of mistakes of course, but also many great moments and I think I look back at that with a lot of good moments and I have to say that also I remember that time as our sort of hypothesis that software would be a zero to one technology for threat forwarding as disproven.
Why?
Ultimately if you think about an operator, let's say doing a CFR shipment, they need about four hours to move something from A to B, let's say from spot rotation or transport order to file closing. And a significant part of those four hours is comprised of working with unstructured data rate cards obviously tons of emails, maybe 50 to 100 to 50 emails depending on how difficult the shipment is between all the parties.
Word files, Word sops process descriptions, Excel reports, etc.
And we found it incredibly hard to move all that unstructured communication onto portals, onto APIs. And I think the reality is that obviously this industry is full of so many nuances, details. The physical industry, of course, things are a little bit different all around the globe. If you go into the details. That was just really hard to cram the reality of this, of this sort of physical logistics world into relational tables and data structures. We also found that at Forto, at my time with software, even where we had the data, it was really hard to automate the decision making in the head of an operator.
I think pragmatically speaking, the power of classical programming, if then else, et cetera, isn't powerful enough to encode the decision making that a human operator can do in her or his hand. And so I moved on sort of with that insight. My first daughter was born three years ago. I wanted to take a break and in that break I handed over my operational responsibilities.
In that break, ChatGPT at its big moment and Genai became really big.
I went back to do some lectures again, do backpropagation on paper, really understand the technology.
I really enjoyed having lots of free time and felt, hey, maybe this is the technology, deep learning, generative AI that I would have wanted to have building out the tech at Forto, because suddenly we can do two things. Obviously process everyone, I think that's super obvious these days, but process all the unstructured data and we have superhuman or going to be superhuman intelligence in a computer that we can use to replicate or maybe even do better human decision making. And that was the motivation to start Zauber, which is the company I'm building today, which where we, we want to build the AI workforce for global logistics, which means we want to be able to orchestrate 80, 90% of what operators currently do with AI.
[00:05:01] Speaker A: And Zalba is the German word for magic. Have I got that right?
Yes, it's a good name. It's a good name. I think it's so impressive that you, that you've founded, you one of the founders of Forto and then you founded this new company and you've been in the game for like well over 10, well over 10 years. It's 11 years, I think, officially, yes. We're going to be today diving into the role of AI and agentic AI in supply chain processes, what the benefits are and how to actually employ these tools. So firstly, kind of broadly speaking, when we talk about AI and agentic AI, can you define what those terms mean as they relate to supply chains. And where does that technology kind of stand today?
[00:05:39] Speaker B: Yeah, agentic AI means that you have a computer system that uses a deep learning large language model to execute a process in a much more free or self determined way than a traditional workflow definition.
Let's say that you know from a robotic process automation or uipath or like classical if then else sort of programming styles where things had to be mapped out in a very deterministic, very clear, very non ambiguous way. And I think we all learned that that is not cutting it for logistics because there's too many exceptions that you would need to describe, too many loops where you would need to go back to some other step, et cetera. And so with agentic AI you're effectively using large language models to execute a process, which means that you give them the ability to work with, talk to existing systems like in Our case office 365, read emails, search emails, write emails, maybe speak to your transport management system, read data, write data, your rate management system, potentially you give it a goal and because the system has sort of a significant amount of intelligence built in, it's able to achieve that goal by executing a series of steps which it can decide freely on. That effectively is agentic AI.
[00:07:07] Speaker A: Right.
[00:07:07] Speaker B: Whereas obviously everyone's using ChatGPT these days. So I think probably most people have also or like any other product copilot, I think most people have a good intuition about where we are going with this technology.
Where is it specifically for workflow executions, workflow execution for operational tasks and logistics.
When I built the first pilots three years ago, I think it was very often models were to dump.
What that means is you would tell them, hey, organize the delivery of a container, reach out to that person, depending on whether they can do it or cannot do it, reach out to someone else or talk to the consignee and then check the time with them, blah blah, blah. Very often the models were too dumb, which meant they wouldn't follow your instructions. You would give them an instruction, but they just wouldn't follow it.
They hallucinated.
That changed quite a bit over the last two to three years.
Models got much more intelligent instruction following.
So hallucinations in a certain way is not a problem anymore.
The costs have come down significantly.
Models work much faster now.
Also, when we started out, getting capacity from providers was not a problem anymore. So I think we are today at a point where we are not only seeing that AI is capable of executing quite a few of the workflows that you do in logistics But I think something else has happened over the last six months which is that an increasing number of use cases that we do is now run on Autopilot. You know, 12 months ago, most people, most deployments that we did, and I think also our intuition strongly was an AI suggests an action to be taken. Let's say an email comes in, a transport order from a customer, document is missing, AI suggests to go back to the customer, he already writes you the draft, hey, this and that is missing. And then you approve it as a human to get sent back to the shipper, the email and you always are in control.
We're increasingly seeing that models do not make mistakes anymore, but also that customers are asking us to turn on autopilot. Because of course if you can stop doing all these nitty gritty task level, human level loop steps, there is first of all a much bigger sort of I think gain with regards to mentally letting go of a specific process chunk. But also things happen much faster because there's a new wait for a human to review things. And I can really drive productivity of our process even further.
And what that means for where are we today to close the loop here is that I think we're increasingly seeing intelligence saturation of AI models or AI for probably what we think around 15 to 25% of the time spent of a threat forwarding operator, meaning I think AI can do that without human supervision already fully autonomously. If you have properly onboarded, integrated it,
[00:10:07] Speaker A: that's a lot of time that it frees up presumably to do other more important things from the conversations that you're having with your customers. What does this tell you about where the freight forwarding market actually is in terms of AI adoption versus where the hype kind of suggests that it might be?
[00:10:25] Speaker B: Yeah, that's a great question. I think that 2025 was the year where I think lots of startups were launching announced big funding rounds still this year, but I think mostly last year that went into this industry and also others with big promises of automation and better customer service and higher productivity, et etc.
I would say this year for me is characterized by at scale companies doing pilots, proofs of concepts, but also the leaders of the our customer cohort doing larger, committing to larger scale rollouts to push the first wave of sort of AI agents into all branches globally.
And I think next year we're going to see that also published on the shareholder investor financial year reports of those first leaders. Whereas I think like most others maybe will follow suit next year with like a larger scale larger scale implementation. I have to say I'm.
I'm positively baffled by how much demand and interest there is at the moment for software or technology in the fret forwarding space. You know, when I started 2015 thinking about building Forto or Fret Hub, how we called it at that time, I met a Fred from Border with my co founder Michael and we showed him our first pilot that I had built at that time, which was creating spot quotes quickly through a platform. He could sort of on a website, enter Shanghai hamburger delivery to Berlin and you'd get a price instantly. And the guy said, oh boys, it's really nice, but no, you can go back to Berlin with your Python code. This is a people's business, you know, it doesn't really matter. Or someone else said digitalization is nice, but it doesn't really matter for our industry.
And I think that has definitely shifted over the last 10 years and even more so the last two years that I think everyone's looking at it and definitely many people are moving on it. And in that sense it's very differentiated from blockchain or predictive analytics or ML in our industry.
[00:12:50] Speaker A: It's really interesting that you say that because something that I hear a lot is that change management is still a major issue. But I'm wondering if that is a bit of a misconception because as I said from what I hear and what you've just said, a lot of freight forwarding companies actually seem quite keen to onboard this kind of technology.
So is change mindset within organizations still a major hurdle or are there any other misconceptions that you've noticed that you kind of want to clear up right now?
[00:13:17] Speaker B: I think definitely change management is still a big topic, a very big topic.
It also seems like the obvious truth if you talk to any AI, vertical AI company at the moment. And I also think change management is an understatement. I think probably we shouldn't call it change management, but rather a transformation.
I think AI will have such profound impact on our industry because you suddenly can be significantly more productive.
You can deliver significantly more standardized but also customer individual service at the same time. Because now you have a technology at your hand that every single time it gets an email from a customer can consider the full SOP of that customer. All the past emails, all the data in your transport management system and all your expert knowledge on how you handle a recent example from one of our customers, pharmaceutical shipments and how we want to do that. Right? We don't route that via Dubai if it's a passively cool shipment because it might be on the tarmac for 45 minutes and then degrade the goods maybe too much.
And so you suddenly are in a position to deliver significantly better services and significantly more individual services to your customers. And so I think it will have profound effects. And I think change management definitely is a big topic.
I think companies have different levels of being able to deal with that.
I think you, the companies that are able to do that best are the ones that have more recently done a bigger rollout of a transport management system, because they still have that muscle of teaching everyone a new tool, looking at your processes and how they match with what you want to roll out, etc. What we've also seen is that I think it really requires leadership to have a strong story how AI will make you a better company with a better outlook in the future because you can serve your customers better and make more money. I think AI can definitely and should come with. And it's like most certainly coming with questions from everyone about, is it going to take my jobs?
What is it going to mean to our business model?
How much is it a threat, how much is it an opportunity? So I think more than ever, much more than for a TMS rollout. You know, switch your AS400 to cargo voice or so. I think now it's the CEO's role to come forth with a really strong story about how AI will help you as an organization to be better off than before and why it allows you to compete even better than not having it.
You asked me about misconceptions.
[00:15:54] Speaker A: Yeah, no, please.
[00:15:56] Speaker B: Should we still go there?
[00:15:57] Speaker A: Yeah, if you've got something.
[00:16:00] Speaker B: I think one thing that came to my mind as you were saying, that's a lot of time about the 25% automation now.
And then you said, oh, that's a lot of time. I think a big misconception that I want to talk about quickly is that many people think about AI as too much of a productivity driver, because very often we speak about time.
When you look at the labs and sort of research, very often now people benchmark AI progress in how long can AI work autonomously, measured in how long would a human spend on that? This morning I listened to Noam Brown from OpenAI, a podcast, and basically he said, yeah, so now we're at a point where sort of models can work a few hours on math problems, maybe soon it's 15 hours and soon it's 100 hours continuously and basically make no mistakes.
And so I think coming from that perspective, I Think we should not fall into the trap of thinking this is only a productivity topic because I think productivity is also, if you only think about productivity, is that definitely a race to the bottom. And my deep conviction also from my own times that running a freight forwarder is.
There is so much yield and lost gross profit opportunity on the table of almost every freight forwarder that I think we should focus much more on that. As I said before, charge every waiting time, be on top of every demerge attention case, have the maybe the specific demerg attention tariff that you discussed with the forwarder in mind. Do allocation manage properly? Because finally you can have the visibility of all your inbound volume really really early and you can push out directives which contracts to use into your operational teams around the world much, much earlier.
But also I think we are able now to with AI put organizations into a position to deliver significantly more individual customer service. I remember from my forte days how every time we closed a bigger account, sort of basically the sales team would come back and say ah, we closed this new account. You know, they just want us to send the report on Friday about all the purchase orders that are coming in into Hamburg with a prefix 1, 2, 3 and the ones with the suffix XYZ. You need to leave the containers there overnight. And by the way, the credit notes need to be sent in yellow. If we are not able to hit the transit times, you need to send a pre alert with four different other options. And by the way, for air freight shipments over a ton, you need to get a green light from the destination and 70 other rules.
And this customization was never really priced in and it was always a burden because it was more work. But also it's really hard to uphold the promise to the customer because maybe that's that branch where the lion's share of that account is being carried, maybe that the people in that branch remember the long SOP for that specific customer. But then you'll have folks in your other branches around the world, in Asia, in North America and South America. And for them, you know, they only do 10 shipments a month on that account and so they didn't even know an SOP exists for that customer. And they completely forget to send any pre alerts when the transit times are not met and that report and so on and so on.
And as I said earlier, I think with AI there's an opportunity to do incredibly individual customer service every single time.
And I think it will become an asset. And I think if you only deliver the standard service, then you will Even I would go so far to say you will not survive as a threat for border because the expectation is going to be in five years from now that you can't do everything that the customers want because you only need to onboard it once into sort of your agentic operations setup. And then obviously again because it's AI, it's 100% repeatable, it will work every time.
[00:19:45] Speaker A: It's really interesting to hear that AI has a role in the customer service side of things, as you said, because that feels like one of the kind of main areas where people say no, that's going to stay very people focused. So it's interesting to see that.
[00:19:56] Speaker B: I think it will stay very people focused. But in the relevant moments when there is something really problematic and the customer is picking up the phone, I think there should be a human on the other side. I think for the QBRs there should be definitely more focused than ever to doing those. Well, I think, you know, our customers will also need to move further and help customers with orchestrating the supply chain end to end, giving them SKU level B2C level shipment visibility and not just move containers from A to B, but finally also I think, you know, they're in a position to be able to do that.
And so I do definitely think that there will be humans very relevant. Humans will be very relevant, but the roles are going to change very significantly.
[00:20:45] Speaker A: It feels to me kind of despite all of these like minor change management things that we were talking about previously, it kind of feels like there's a certain inevitability around. I think a lot of companies now know that AI is something they're going to have to get on board with one way or another, if nothing else, just to keep up with their competitors. So I think the big question is kind of how to go about it. And so for a forwarder looking to introduce AI or Agent Ki or this kind of technology, how should they think about the build versus buy question? I mean, what are you seeing companies do today and where do they often get that decision wrong?
[00:21:18] Speaker B: Yeah, how to go about it? I think the first thing is not even, you know, answering the build versus buy question in the first place. But don't do a strategy project, maybe even with a consultancy for six or nine months and discuss the theoretic opportunities, get the exec team to use Claude code or, I don't know, lovable or replit and build an intuition and a deep understanding of what AI really means. You know, many people say AI is superhuman, but it also is in some weird Areas not superhuman at all. Intelligence of AI is very jagged. When it is really capable it's 100x superhuman.
But in some areas it's absolutely not. And I think you need to build and intuition and you need to build that conviction yourself. I think it's really hard to get that conviction on slides, first of all. Second, when you think about building it yourself or buying it, I would argue that, well, I'm still a big believer in Adam Smith. I think, you know, I think just there's just significant comparative cost advantages when you're able to focus on a single thing that multiple people want. And I think that's also the reason why this industry exists because you know, some people focus in producing certain things in one part of the world, but they can also maybe do it more cheaply and other things can therefore free up time and specialize in some other different value generating activity.
If I were the CAO of the Fred for border, I'd definitely try to get my hands dirty myself. But I would do it in areas where I think there's outsized strategic value creation because I have a specific opinion or right to win, right to play as opposed to my competitors that might be.
I'm already much closer to customers with regards to order management looking at SKU purchase order level and so I probably want to build some kind of intelligence platform visibility in that topic. Maybe you have a very particular opinionated way to deal with my carrier contracts and do allocation management. So I should probably spend some time on that.
But I also think that, you know, it is probably not possible for every company to do everything yourself, despite coding getting certainly easier.
And you know, my pitch would probably be for now our company, but I think our competitors would say the same.
We've seen hundreds of thousands of transport orders, so initial customer bookings go through our system. We've seen all kinds of documents, hiccups, big documents, small documents, all kinds of formats, glitches, edge cases. We've been able to tune the cost of that use case to such a low level and the way that honestly fretful order A accepts a booking versus fretful Water B accepts a booking is not so different. So there is some rather commoditized aspects of your value creation pipeline that you just want to forget about and get someone that can get you live in a few weeks and focus again on the more higher value activities.
[00:24:31] Speaker A: How do you think this build versus buy question has shifted over the last 10 years? Because obviously you started forto in 2015, which was 11 years ago. You've Been very busy since.
So you've seen many freight forwarders at the beginning of their technology and AI adoption journeys. So are there any kind of patterns you've noticed of where they start and what they get right and what they get wrong? I mean, have these stages of adoption changed? Because presumably now the capabilities of these companies are quite different. I feel like 10 years ago they probably didn't have much choice to build their own.
[00:25:01] Speaker B: Interesting. I think that.
After everything that's happening in the TMS world on the threat forwarding side, probably there is a set of companies that have built their own transport management systems that are now really happy about it because they have sort of no dependency on a player that would be able to increase prices, lock them out of their own data, etc. On the other side, I think you also have to ask yourself the correct question.
Would you have been able to build that system of record yourself?
Would you have been able to sort of also own it throughout the ups and the downs of the industry, keep it updated, etc. And I can definitely say that there is one pattern among our customers which is that yes, no one loves their system of record, but if you have a standard system of record which is not homegrown, no matter how bad the API of that system is or the edi, it at least exists. And very often it exists at such a large depth that you can do 95% of the things a human operator can do through an EDI or API. And that is really important for AI because if you do not allow AI through an integration technically to have the same option, space, action, surface than a human, it's going to be impossible to automate anything exciting. Obviously you can't say I have hired a super smart, Einstein, genius freight forwarding clerk and now they can only look at bookings and maybe create a milestone in the system, nothing else really is possible.
[00:26:47] Speaker A: Right.
[00:26:47] Speaker B: That will not work out. The person's not going to be productive at all.
So there is, I think both things are true at the same time and I think it's with again software building software becoming more democratized, again a discussion that people are having. The jury is out. Let's see. I, I think the next three, four years are going to shed light on that question.
[00:27:14] Speaker A: Well, it's something that I hear that's quite big challenges for freight forwarders. I mean there's quite a lot of noise around technology and agentic AI. I mean it isn't a secret that there are a huge number of companies offering AI solutions in this industry. So how do you differentiate Zalber from the growing number of AI claims and probably products in the market? And how do forwarders know who to partner with?
[00:27:38] Speaker B: I think, you know, everyone needs to make up their own decision. I'd say very simply, I've run a threat forwarder for 10 years with I think relevant scale.
So I think it's not only true for me, it's true for 50% of the team here. We deeply understand what threat forwarding means, where the problems are, where the value is being created, how those companies work.
And I think that really, really matters when you want to bring the power of the technology on the road.
We, I think have a good product, but I think everyone needs to think about that and make their own judgment. And I think we see very, very positive feedback from our customers. We Today work with seven of the 25 largest Feders in the world. So I think there's some positive indication. But again, I think everyone can form their own opinion.
[00:28:30] Speaker A: This episode's gone really quickly already. We've already been talking for well over 25 minutes. I just want to kind of wrap up by looking to the future and kind of asking you for a forward looking statement. So how do you think AI is going to fundamentally change the freight forwarding industry over the next three to five years? Because there's quite a lot of excitement around generative AI and we've touched on a few of the reasons for that. But beyond the kind of obvious promises of automation and efficiency, where do you expect to see the biggest AI driven changes in the freight forwarding value train chain? And which areas do you think will be kind of much harder to transform than people might expect?
[00:29:07] Speaker B: I think, you know, overall I think you're going to see a stronger bifurcation because I definitely believe in the power of the technology to significantly transform the business model into a higher value, higher, higher margin sort of business than it is today. And so I think there's companies that are going really hard for it and there's companies that maybe don't have the capacity or capital to do that. I think probably even especially the longer tail of the market really needs to make sure that they can also leverage the power of the technology as well as the medium sized and big sized companies.
So I think there's a stronger bifurcation between people who adopted or adopted fast and have this change management muscle and the ones that have not. And I think it's going to Show Definitely in EBITs, it's going to show at a higher level in volume retention, customer retention rates. And I think we'll see the first results of that next year. Already?
[00:30:00] Speaker A: I mean, it's so. It's so fast moving. I feel like it's quite mean of me to ask you to predict something that might happen in five years.
[00:30:06] Speaker B: We should just talk again next year.
[00:30:07] Speaker A: Yeah, exactly.
Good idea.
[00:30:10] Speaker B: What we've.
[00:30:10] Speaker A: What we find for joining me. It's been wonderful to speak to you.
[00:30:13] Speaker B: Thank you very much, Charlotte. Thank you for having me.
[00:30:17] Speaker A: Wonderful. Okay.
