The orchestrator: Automating the global supply chain with Josh Clement-Sutcliffe, Zencargo

Jim Clark: Josh, it’s an absolute pleasure to have you on this call today. You are AI and Automation Product Manager at Zencargo. What does that role involve?

Josh Clement-Sutcliffe: It involves a lot and it feels like it’s constantly shifting. The fundamentals of a product manager role are still the same — being the voice of stakeholders, understanding business objectives and priorities, coordinating many different departments and teams to make sure we’re all moving in the same direction. The change I’ve experienced over the past year is how AI and automation are now starting to play a major role internally and within the supply chain industry, and having to adapt and move so quickly with these technologies. That’s the biggest change. But fundamentally, the core of what it is to be a product manager is still the same.

Jim Clark: Tell me about Zencargo’s North Star when it comes to AI.

Josh Clement-Sutcliffe: Our Road to 2030 is to automate the supply chain. We will have an autonomous supply chain by then. Fundamentally, to achieve that means we need to be removing manual steps within the supply chain process — that’s how we’re looking at reducing friction. We’re looking at reducing bottlenecks, the amount of time it takes humans to action a specific workflow or the next step on a particular shipment. By moving towards our North Star of automating the supply chain, we can get into a world where operators are just overseeing what AI and automation is doing, starting to move into a world of handling relationships and managing exceptions, rather than doing more admin tasks, which has been the case in the past.

Jim Clark: It sounds like you’re stitching pieces together to make that vision a reality. Where are you on the journey?

Josh Clement-Sutcliffe: The goal for 2025 was to prove AI and automation within the supply chain context. We know these technologies work, but we’re very much at the forefront of actually influencing them within our industry. 2026 is really making sure we have the guardrails and the operational processes and maturity in place to make sure they can scale — with the visibility and human element we need to make that a reality. We started last year really defining the areas where AI and automation can have the most value. We’re now taking those areas where we’re seeing the most impact and tying them into our orchestration layer — Luca — which has multiple different AI products feeding data into it. Later in the year we’ll start to expand Luca’s outputs into different parts of the business as well.

Jim Clark: Where have you seen the greatest success so far?

Josh Clement-Sutcliffe: Really twofold, and this is where we’ve been doubling down in terms of our strategy. AI and automation is only ever going to be as good as the data it’s built on. So we made the strategic decision to really focus on that area last year. The biggest impacts have been in what we’re calling our LUCA Scan — our document parser product. This is us training and fine-tuning models to work within the supply chain specific context. Document parsing might not sound the most interesting, but when you’re really working across so many different document types, all languages, across the globe, across different standards, it becomes very, very challenging. It’s not just structured receipts in the same format. We’re also tackling unstructured data from emails — all third-party communications from our suppliers and customers. Every time those emails come in with unstructured data, we’re structuring it, sending it into our orchestration layer, and then getting downstream automations on the back of that. The ingestion layer of our orchestrator is what will ultimately empower it to automate the supply chain.

Jim Clark: A lot of mid-market firms realise the foundational work should have happened before they started building automation and agents. Was this a deliberate process for you?

Josh Clement-Sutcliffe: I’d say we knew what we needed to do because of the errors and mistakes we made along the way. We also fell into that trap. From August last year, we made the strategic decision that this is what we learned. If we’re going to make a success of this, this is how we’re going to make it a reality. And so we put a transformation plan in place. Data ingestion, hygiene and transformation was just the first step of that. We also fell into some of the same pitfalls, but once we recognised the mistakes we made, we made a plan to rectify them.

Jim Clark: Are there any instances that jar on your memory — things you wish you could go back and change?

Josh Clement-Sutcliffe: Yes, absolutely. There were a couple of projects we worked on probably for longer than we should have done. I think that’s natural — you fall into the sunk cost fallacy at that point. But that’s part of the learning process. These are new technologies. Creating a playbook in this industry for AI and automation, which hasn’t been done before. So these mistakes are going to be made. It’s not so much the fact that there are mistakes — it’s your ability to pivot and to learn from them. I think we’re on the right path now on the back of those.

Jim Clark: Tell me about the LUCA orchestration layer. What is it and why did you build it internally?

Josh Clement-Sutcliffe: This is essentially the brains of what we’re doing with AI within Zencargo. LUCA is the AI product that’s going to define our success in that North Star. This is where we’re working with models and fine-tuning them so that AI can understand what needs to be done in a shipment context. For instance: a booking comes in saying we need to get goods from somewhere in China to London. Once we get that booking — let’s say it came through via email — we need our orchestration layer to figure out the intent. AI needs to see that email and determine: this person has requested a booking. To progress this, LUCA needs to understand the intent, go to the booking service, get a response back. If there’s a problem, LUCA would figure out it needs to send an email and request more information. Or it can continue the automation from there. This is how we’re going to transition our staff from daily admin tasks to supervising and overseeing the automation within the supply chain.

Josh Clement-Sutcliffe: We built it internally because of the volatility of this industry. There are so many different parties and so many different variables involved in the supply chain process that so many things can go wrong — and completely unforeseen things can go wrong. The most famous example is the container ship blocking the Suez Canal, where you have to divert all containers around Africa rather than through the Mediterranean. That’s obviously a one-off example, but these are the real-world issues we have to deal with. Because of this complexity, we felt there was nothing in the market that would have the supply chain context to understand and react to these situations. We had to make the decision that this was something we’d have to do internally.

Jim Clark: You’ve mentioned taking people on the journey with AI. How do you manage that cultural change?

Josh Clement-Sutcliffe: We’re very fortunate to have the full support and backing of management. One of our goals this year is to be known as the AI freight forwarder. We’re differentiating ourselves within the marketplace based on how we work with AI. That isn’t just about the products and technologies — it’s very much an organisational shift. One of our KPIs internally is to make sure a certain percentage of departments are using AI in their workflows, and the percentage of people in the business who are using AI. That buy-in from management makes my job a lot easier. But the problem of getting people trained, making sure they’re up to speed, they know what AI and automation is doing — that’s of course still a challenge. And those are still core problems that product managers have to engage with. Human nature isn’t going to change because of these technologies. We need to make sure we take people on the journey. AI can be quite scary. We need to be clear we are not trying to replace jobs. What we’re trying to do is allow operators to be spending their time on the high-value issues rather than just the laborious manual tasks and admin.

Josh Clement-Sutcliffe: We’ve accepted that we can only get to a certain level of success because of how AI is fundamentally built. We accept that errors will take place. But we’re building in auditability by design to make sure errors can be picked up as early as possible and problems mitigated. These are very much problems that have existed in software and technology for the past 30 years. I don’t see these as inherent to AI specifically — it’s just a human problem. We need to make sure we’re doing the correct change management to bring the entire organisation along.

Jim Clark: How has management support made a tangible difference?

Josh Clement-Sutcliffe: They’re setting the direction of the organisation. The expectation internally is that to become the AI freight forwarder, we need to be leveraging these technologies across the board. They’re very active in making sure we have everything we need to be able to move at pace. When they’re required to step in, they will do so, and they’re very hands-on in doing so. Giving teams the autonomy to make decisions on direction and how to achieve the organisation’s goals — they do a good job of recognising that. But should the need arise to step in and push things along, they absolutely get involved.

Jim Clark: You mentioned before that a lot of your operators were doing quite tedious copy-and-paste work. How have they responded to seeing that go away?

Josh Clement-Sutcliffe: Absolutely positively. And I think that’s one of the good things about the approach we’ve taken — it shows very quick benefits. For the email project, by taking unstructured data, making it structured and putting it directly onto the platform, people are seeing the things they had to do themselves done for them. Copying and pasting — that’s done. Attachments taken directly from emails and uploaded to the correct shipment — done. Our LUCA product is already starting to make suggestions based on that data, both corrections and potential optimisations. They’re now starting to feel the benefit of it. It becomes easier when it moves away from being theoretical and starts to show actual impact. And being able to actually trust that impact — being able to see that it’s working to a certain degree and feel like you can step in if you need to — is crucial.

Jim Clark: You mentioned the four stages of your transformation journey. Can you walk us through them?

Josh Clement-Sutcliffe: The first stage is data foundations and ingestion, which we’ve talked about a lot. The second stage — and one of our major initiatives for 2026 — is visibility and trust. We need to take into account the ways operators actually use the platform and make sure we’re surfacing the information they need at the right place at the right time. These are best-practice UX principles — they haven’t really changed. What has changed is the amount of potential noise we could be sending to operators. One of the lines we’re treading carefully on is how we define what’s useful versus just overwhelming people with information. That’s a challenge we’re currently fine-tuning. The third stage is partnership — when we’re really starting to automate predictions, risks and optimisation capabilities and operating at true scale. But we need to get the visibility and trust element right first, because otherwise if we start to automate actions beforehand, we’re just going to be automating mistakes.

Jim Clark: Have you changed how you recruit in response to AI?

Josh Clement-Sutcliffe: Yes, absolutely. Our HR team is changing the tools and systems they use, so AI is happening in that department too. But we’ve also started to change our hiring process and the questions we ask candidates about AI. We now want people to show us what they use with AI. If a developer is using AI to build a solution in a technical test, we encourage that. We want them to use AI. The question isn’t whether they use AI or not — it’s really making sure they have the skills and understanding to know exactly what’s going on under the hood. We encourage it, we want it to happen. We just need to make sure people have the knowledge and skills to back up what they’re actually delivering.

Jim Clark: If a product manager said: I want to start automating with AI but I don’t know where to begin — what one piece of advice would you give?

Josh Clement-Sutcliffe: I understand, because I was in a similar position last year when I started at Zencargo. The shift to leveraging AI and automation is both a technological and a transformational shift. The transformational shift will be a lot easier if you can prove the value of using AI and automation in any given area first. When we were looking at how we were going to leverage these technologies last year, when it became obvious which areas would have value, there was definitely a light-bulb moment — we thought: we could actually be moving into a whole new world if we doubled down in this area. So I would just say: look for those areas, play around with it. You don’t have to get it right. And when you find something that could be really powerful, you will get that light-bulb moment. If you can prove that value, that makes the transformational shift internally so much easier. Start small. If you can see some value in something, follow that path rather than going down too many different routes at once. If you find the area where it’s powerful, you will know it.

What is Zencargo’s Road to 2030 and what does an autonomous supply chain actually mean?

Zencargo’s Road to 2030 is the company’s strategic goal to fully automate the supply chain by the end of the decade. In practice, this means removing manual touchpoints from the shipment process so that operators shift from doing administrative work to overseeing what AI and automation is doing — managing relationships and handling exceptions rather than processing data. Josh Clement-Sutcliffe describes this as a fundamental change in how operators interact with freight and supply chain, not simply a cost-saving exercise.

What is LUCA and how does Zencargo’s AI orchestration layer work?

LUCA is Zencargo’s proprietary AI orchestration layer — the intelligence layer that connects multiple AI products and automates decisions within the supply chain. It is designed to understand the intent behind incoming communications, trigger the appropriate workflows, request additional information when needed, and progress shipments with minimal human intervention. Zencargo built LUCA internally because the volatility and complexity of global supply chain logistics — with its many parties, variables, languages and standards — made off-the-shelf solutions inadequate.

Why is document parsing so important for AI in logistics?

Document parsing is the process of training AI models to extract structured data from unstructured sources such as emails, shipping documents, invoices and certificates of origin across different languages, formats and standards. In global supply chain operations, this is one of the highest-leverage automation opportunities because so much communication still happens via email and in formats that vary enormously by country, supplier and document type. Josh Clement-Sutcliffe describes it as unglamorous but foundational: if the ingestion layer cannot reliably structure incoming data, nothing downstream can be automated reliably.

What mistakes did Zencargo make early in its AI journey and what did it learn?

Zencargo fell into some of the same traps it now helps others avoid — working on early projects for longer than was justified, and trying to automate the ‘happy path’ of workflows without first solving the data foundation problem. The key learning was that AI and automation are only as good as the data they are built on. From August 2025, Zencargo put a formal transformation plan in place with data ingestion, hygiene and transformation as the first step. Josh Clement-Sutcliffe is direct that mistakes are a natural part of building something that has not been done before in a given industry — what matters is the ability to pivot and learn from them.

How should organisations manage the human element when introducing AI automation?

Josh Clement-Sutcliffe’s approach is built around auditability by design: ensuring operators have a full audit trail of everything AI is doing, a clear human-in-the-loop element, and the ability to step in when something goes wrong. He is explicit that something will always go wrong — that is how AI works — and the goal is not to eliminate that possibility but to detect and mitigate errors as early as possible. Critically, he frames automation not as a job replacement exercise but as a way of repositioning operators from labour-intensive admin into high-value judgment work. This framing has made it significantly easier to bring teams along with the change.

When should a mid-market business build AI tools in-house versus buying from a vendor?

Zencargo’s default is to look for an off-the-shelf solution first. But for its core supply chain intelligence — where the complexity, volatility and specificity of global logistics made no existing vendor solution adequate — it built internally. Josh Clement-Sutcliffe applies a straightforward principle borrowed from Occam’s razor: if a simpler solution exists that fits the need, take it. Where third-party tools are used, Zencargo evaluates them rigorously on whether the data they provide is accessible in the format and quality needed to feed downstream automation. The question is not just whether the tool works in isolation but whether it will accumulate problems in the long run.

What should product managers new to AI automation focus on first?

Josh Clement-Sutcliffe’s advice is to start small and look for areas where AI can provide immediate, demonstrable value. The light-bulb moment — when a team realises they could be moving into a whole new world if they doubled down on a particular use case — is what makes the subsequent transformational shift manageable. He warns against spreading effort across too many different routes at once. Once you have proven value in one area and the organisation can see and trust the impact, scaling becomes far easier because you have real evidence rather than theoretical arguments.