Walk into the back office of a mid-sized freight forwarder anywhere in Europe and the scene is the same: two monitors, email on one, a transport management system that may be older than the person sitting in front of it on the other, and a team of operators spending their day copying data between them. A year-old Munich startup called 5U AI closed a $3.2 million pre-seed round today to replace that copy-paste loop with stateful AI Workers that quote, book, track, and reconcile invoices across air, sea, and road freight — and, critically, remember every decision they make along the way.
The round was led by London-based Emerge Capital, a $75 million pre-seed fund known for backing future-of-work and human capital development startups. Co-investors are a cohort of senior logistics executives from DHL, GEODIS, DSV, Maersk, and Ceva Logistics — operators who have spent careers inside the systems 5U AI is automating, backing the company with personal checks rather than institutional capital. At pre-seed stage, that is an unusual level of domain validation.
The funding marks 5U AI’s first public financing since its founding in late 2025 by Yagiz Abik (CEO) and Fehmi Şener (CTO), both graduates of the Technical University of Munich.
An Industry Running on Two Screens and a 20-Year-Old System
The freight forwarding industry coordinates the movement of goods for manufacturers, importers, and exporters across ocean, air, and road carriers — a global market projected to reach $235 billion by 2026, growing at more than 5% annually. The software layer on top of it is far smaller but growing fast, from roughly $530 million today toward an estimated $1.7 billion by 2035.
Despite that scale, the industry’s back-office operations are stubbornly manual. Abik has described visiting freight forwarder offices across Europe before founding 5U AI and finding the same workflow everywhere: two screens, email on one, a legacy TMS on the other, and operators copy-pasting between them for hours. Up to 70% of forwarding work is repetitive in exactly that way, according to 5U AI — a figure the company bases on discovery calls with operators, though it has not been independently audited.
Prior attempts to close that gap have largely failed. General-purpose automation and robotic process automation (RPA) platforms, which follow rigid if-this-then-that rules, break whenever a carrier portal changes its interface or a rate source goes offline — a common occurrence in an industry with hundreds of fragmented carrier systems. Industry research puts RPA project failure rates at 30–50%, with 45% of firms reporting weekly bot breakage that requires manual intervention. General-purpose AI chatbots fall at a different hurdle: they are stateless. They answer a question and forget the session, which makes them useless for shipment threads that may go dormant for three months before a customer resurfaces asking for proof of delivery.
Why Stateful AI Workers Differ From Chatbots — Technically
The distinction 5U AI draws between its AI Workers and a chatbot is not primarily a marketing claim — it is an architectural one. A chatbot is stateless: each query is independent, with no memory of prior interactions. 5U AI’s Workers are stateful: they maintain a persistent context across the full lifecycle of a shipment, from initial quote request to final invoice reconciliation, regardless of how much time passes between touchpoints.
Under the hood, the platform runs a combination of frontier models — large general-purpose language models for complex reasoning tasks — and smaller, fine-tuned open-source models for narrow, well-defined freight operations. The key technical finding the company is building around: a small, specialized model often outperforms a large general-purpose one on tightly defined tasks like reading a carrier rate sheet, extracting shipment data from a messy PDF, or classifying what a multi-party email thread is actually about. That advantage holds specifically because fine-tuning on domain-specific data can produce a model that has deeply internalized a narrow task, while a large general-purpose model must infer the same task from context on every call.
The training pipeline has two stages. First, industry professionals annotate real freight cases to create ground truth — what the correct quotation, booking, or invoice match actually looks like — which is used to fine-tune the base models on the logistics domain generally. Second, before any AI Worker goes live with a specific customer, it is fine-tuned again in “shadow mode” on that customer’s own historical cases: their specific margin rules, carrier preferences, and document formats. The model runs alongside human operators without taking any action, absorbing their institutional knowledge, before being allowed to work independently.
Every model update then runs through 5U AI’s internal evaluation platform, called Grader, which simulates thousands of freight scenarios — an expired surcharge, a customs broker joining a thread mid-conversation, a dangerous goods request that must be escalated — and scores the outcome before anything touches production.
What the Decision Layer Does — and Why It Matters to Buyers
The most architecturally significant element of the 5U AI platform is what the company calls its Decision Layer (also referred to as the Context Layer). Every action an AI Worker takes is captured as a structured record: what it decided, on what data, and with what confidence. When a Worker prepares a quotation, it logs the reasoning — which carrier rates it pulled, which margin rules it applied, which customer preferences it honored.
In the early days of a deployment, the Decision Layer feeds a human approval flow. The AI Worker prepares a quotation and sends a message to the operator in Microsoft Teams or Slack: “Here’s my pricing — can I send it?” The operator approves or corrects. When a correction happens, it is written into the Worker’s memory immediately. In one documented instance, a Managing Director questioned why the AI was calculating margin a certain way; the Worker responded by asking: “Your operations lead taught me this rule. Do you approve changing it?” — a behavior that reflects genuine persistence of context rather than a reset on every interaction.
As the system builds confidence, the approval step is removed for routine cases. Live customers are reaching 85–95% automation rates on individual use cases, according to 5U AI — a company-stated figure without independent audit, but one that Abik has discussed in detail with named industry publications.
Freight buyers evaluating the platform should be aware of a structural implication this creates: as the Decision Layer accumulates a forwarder’s margin rules, carrier preferences, and institutional knowledge over months and years, that proprietary data corpus becomes increasingly specific to 5U AI’s platform. Switching to a different system later means leaving behind accumulated decision data that the industry has never systematically captured before — and which 5U AI’s founders have explicitly identified as their long-term strategic asset and a “data product that will benefit all customers over the long run.”
The Self-Healing Architecture
One of the documented failure modes for automation in freight is brittle dependency on external systems. Carrier portals change their interfaces. Rate sources go offline. Two systems disagree on a shipment status. In 5U AI’s architecture, when a Worker encounters one of these failures mid-task, it does not stop or produce an error: it finds an alternative route to complete the job, updates its own memory with the new method, and proceeds. Only when it genuinely cannot proceed does it escalate to a human operator — and it does so with full context, so the human is not starting from scratch.
This self-healing behavior is the architectural answer to RPA’s primary failure mode. Where an RPA bot breaks silently when a carrier portal changes its UI and must be manually repaired by a developer, 5U AI’s Worker detects the change, finds another path, and writes the solution into its own memory so the same failure does not recur.
Operational Results and What’s Been Claimed
The quantitative claims 5U AI has published are striking, and worth stating precisely with their caveats. A typical mid-sized European forwarder receives 200 to 300 quotation requests daily, handled by a few staff members, with many going unanswered. A human operator requires roughly 15 minutes per quotation; a 5U AI Worker requires approximately two minutes and handles multiple requests in parallel. Requests arriving overnight are quoted and won while staff are offline — Abik has said the company sees more than doubled win rates within weeks at selected high-volume customers. These figures are company-stated and have not been verified by an independent auditor.
5U AI’s first named commercial customer is TCI International Logistics, working across air and ocean freight.
A Crowded Race, With a Structural Differentiator
5U AI is entering a field that already has well-funded competitors. In London, Nexcade — founded by Dan Bailey (former COO of supply chain platform Sedna) and Tasho Kjosev (former Technical Lead at Palantir) — raised $6 million in seed funding led by Project A Ventures in July 2026, working with clients including XPO, Zencargo, Cardinal Global Logistics, and CargoTrans. In Barcelona, Opereit raised $2.5 million to address post-transit cargo losses.
Nexcade’s approach focuses on specific agent types — quoting, booking, customer service, pricing — leaving decisions with human operators throughout. 5U AI’s bet is different: the company is building toward higher automation rates by keeping the Decision Layer as a shared memory substrate across all Workers, and betting that the proprietary decision data that accumulates there becomes a compounding moat rather than a feature.
Whether that bet holds will depend on execution in the exception-heavy part of the job — the scenarios that fall outside documented playbooks. “Freight forwarding has many exceptions that are hard to automate because there is not enough documentation,” one industry analyst observed in a 2026 assessment. Abik acknowledges the same challenge: the evaluation platform’s job is precisely to stress-test edge cases before they reach production. 2026 is also the year Gartner predicts agentic AI moves from experiment to deployment at scale — and the year the same firm warns that 40% of agentic AI projects will be canceled by 2027, citing escalating costs, unclear ROI, and inadequate risk controls.
Where the Money Goes
The $3.2 million will fund three priorities, per Abik: scaling go-to-market across Europe (the company has spent zero on marketing to date, relying entirely on direct outreach); deepening R&D investment in specialized inference and fine-tuning; and expanding the Decision Layer into a durable data product. Headcount growth is planned across product, engineering, operations, commercial, and customer-facing functions.
The company’s origin is worth noting: Abik and Şener met organizing weekly amateur football matches at TU Munich — Şener ran a 150-player amateur football community in Munich, with matches every week — and built 5U AI from a thesis Şener had written in collaboration with CEVA Logistics on machine learning applied to road freight. Şener left a BMW engineering role to co-found the company. The founders’ explicit thesis is that deep industry empathy, not just technical capability, is the deciding variable in whether AI agents succeed in a domain as relationship-dependent and exception-heavy as freight.
Frequently Asked QuestionsHow does 5U AI’s platform differ from earlier freight automation tools like RPA?
Robotic process automation follows predefined if-this-then-that rules and breaks whenever a carrier portal changes its interface or a rate source goes offline — a documented problem that causes 30–50% of RPA projects to fail and requires constant developer maintenance. 5U AI’s Workers use fine-tuned language models that reason through tasks rather than follow scripts, self-heal when external systems change, and maintain persistent context across a shipment’s full lifecycle. The architectural difference is that the system adapts; it does not break.
What is the Decision Layer, and should freight forwarders consider the long-term implications before adopting it?
The Decision Layer is 5U AI’s persistent record of every action an AI Worker takes — what it decided, on what data, and with what confidence. Over time it becomes a structured log of a forwarder’s institutional knowledge: their margin rules, carrier preferences, and operational playbooks. For buyers, this has a dual nature: it makes the system smarter with every shipment processed, and it is also the basis for compounding vendor lock-in. As that accumulated data grows more specific to the platform, migrating to a different system later means leaving behind institutional knowledge that has never existed in portable form before. Sophisticated buyers should ask what data portability looks like before committing.
What agentic AI freight tasks can the platform automate, and what remains human?
5U AI’s Workers currently automate quoting (pulling carrier rates, applying margin rules, preparing and sending quotations), booking, shipment tracking, data entry into the TMS, and invoice reconciliation across air, sea, and road freight. The system is designed to escalate to a human when it encounters a genuinely novel exception — a dangerous goods request, a complex customs scenario, a dispute that requires relationship judgment. Live customers are reaching 85–95% automation on individual use cases as confidence builds, with the remaining cases handled by human operators who receive full context from the Worker.
What does the broader freight AI market look like right now — is this a proven category?
Agentic AI in freight is transitioning from proof-of-concept to production deployment in 2026, but it is still early. Gartner predicts 40% of enterprise agentic AI projects will be canceled by 2027 due to cost overruns and unclear ROI. Forrester’s June 2026 assessment found that while 75% of enterprises had adopted agents in some form, only a fraction were running them in genuine production. In freight specifically, the primary challenge is exception-heavy workflows where documentation is sparse. Companies that solve for that — through customer-specific fine-tuning, persistent memory, and rigorous simulation testing before deployment — are ahead of those deploying generic AI tools.