Live Coverage: 'AI Technologies in Intermodal Freight Transport' Webinar

RAILTARGET brings you live coverage from today’s webinar on artificial intelligence (AI) in intermodal freight operations, organised by UIRR under the Combined Transport for Europe (CT4EU) campaign.

The session explores how AI-driven innovation is transforming Combined Transport, from automated processing systems and predictive analytics to image-based safety tools and optimisation algorithms that enhance efficiency and reliability across the CT chain.

14:00 Eric Feyen, Technical Director at UIRR, opens the webinar by welcoming over 300 participants and introducing the topic of AI technologies in intermodal freight transport. He outlines UIRR’s role as the representative body for Europe’s intermodal sector, covering 50–60% of the market, and highlights its mission to promote Combined Transport and influence EU and national policy.

14:05 Feyen explains that the session takes place under the Combined Transport for Europe (CT4EU) campaign, aimed at bringing together logistics providers, innovators, and shippers to advance digitalisation and innovation. He stresses the importance of AI in creating new efficiencies within the sector. He cites findings from Stanford University’s AI Index Report, noting that while digitalisation varies across regions, AI is becoming more efficient, affordable, and widely adopted, with governments introducing new regulatory frameworks to ensure responsible use.

Feyen continues by noting that governments and EU institutions are now developing specific regulations on AI, referring to the European AI Act as a major step toward responsible deployment. He mentions that Eurostat already tracks AI-related data, showing a clear rise in adoption between 2023 and 2024, especially among larger enterprises.

He adds that this trend shows why UIRR considers AI crucial for intermodal transport, as it can optimise processes across the Combined Transport supply chain, from terminal operations to rail coordination and data integration. Feyen concludes by inviting Dimitro Blomme, CEO of Clevermint, to begin his presentation.

14:10 Dimitri Blomme, CEO of Clevermint, opens his presentation with a provocative question: "What if the future of intelligence is not artificial, but augmented?" He invites participants to rethink AI not as a replacement for humans, but as a tool that enhances human capability and decision-making. Blomme stresses that AI is less about technology and more about rethinking how businesses create value and make decisions. The real opportunity, he says, lies at the intersection of strategy and people, where human judgment and machine precision work together. He says that goal is not automation for its own sake, but the augmentation of human intelligence, using AI to empower, not replace, people.

14:15 Blomme notes that to truly benefit from this collaboration, we must first understand what AI really is. Contrary to popular belief, AI is not a single technology but a family of systems that learn and act with purpose. At its core, AI recognises patterns in data to make predictions. He distinguishes between machine learning and generative AI, the latter powering tools like chatbots that can generate text, images, or code.

Automation, he explains, is about speed, efficiency, and cost reduction. AI takes over repetitive, high-value tasks. In logistics, this translates to predictive routing, which helps reduce delivery delays and fuel consumption while making operations faster and smarter.

The second lever, insight, transforms data into intelligence. AI algorithms, as Blomme notes, can detect diseases in healthcare, predict demand in retail, or optimise inventory, showing how data-driven decisions improve clarity and performance.

Finally, augmentation is about enhancing human talent. Developers, for example, use AI copilots to generate or review code, while marketers rely on AI tools to brainstorm and refine campaigns. Blomme stresses that the goal is not total automation but an intelligent balance between human creativity and AI precision.

He illustrates this across industries: AI personalises retail experiences, prevents downtime in manufacturing, supports customer service agents, and in logistics, it optimises route planning, cuts emissions, and improves delivery reliability.

14:25 Blomme compares AI adoption to agile development, where progress comes through iteration and collaboration. Citing the MIT framework on collective intelligence, he explains that the future of high-performing companies will depend on how well they design cooperation between people and AI.

Using a humorous example, Blomme shows an AI-generated compass with five cardinal points, underscoring the need for human oversight to challenge machine outputs. He presents with a four-step roadmap for AI transformation:

• Clarify the 'why' – define purpose, governance, and expected value.
• Build strong data foundations – quality, security, and accessibility are key.
• Upskill teams – make AI literacy a core competence across the company.
• Start small, measure, and scale – test pilot projects, expand what works, and learn from failure.

He wraps up his presentation with a powerful message: "As AI becomes more powerful, leadership must become more responsible. With great power comes great responsibility."

14:30 Serge Schamschula, Head of Partner Management at Transporeon, takes the floor to present AI use cases in logistics, shifting the discussion from strategy to practical applications. He points out that while AI is not new, its public prominence has surged since the rise of ChatGPT, which triggered massive investment and media attention. However, he cautions against following the hype blindly, choosing instead to focus on real, operational AI use cases already delivering value in logistics.

His first example is autonomous procurement, an AI-driven solution that enables shippers to push spot loads to transport providers automatically. The system delivers average cost savings of 10% while cutting human workload by 80%, replacing teams of eight with just one operator. He adds that another Transporeon feature allows the system to process and respond to freight tenders within two seconds, regardless of whether data arrives in structured digital form or as unstructured information from emails. This capability enables logistics providers to handle huge volumes of offers efficiently, improving their chances of securing optimal transport matches with minimal manual effort.

14:40 Schamschula continues by showing how AI implementation at Transporeon has delivered measurable results. By combining autonomous procurement for shippers with autonomous quotation tools for carriers, the company achieved 80% year-on-year growth in its AI-powered product portfolio, which is a clear win-win for both sides of the logistics chain.

He then presents a second use case: AI-based carrier vetting and onboarding. The system automatically performs tax, licence, and company register checks, and even reads insurance documents in different alphabets, such as Cyrillic, to verify their validity. However, Schamschula stresses the need for human verification, alike Dimitri Blomme’s earlier point.

Schamschula notes that across logistics, AI assistants excel in speed and data processing, capable of handling tasks far faster than humans, but warns that their performance depends entirely on data quality. Without strong knowledge bases and reliable datasets, "AI is completely lost."

He adds another example from Transporeon’s own operations, AI-assisted software coding, which now helps developers generate and review code more efficiently. As this capability expands, he predicts that AI-supported software development will significantly reduce costs across industries.

14:45 Schamschula moves to his closing reflections,.

First, he asks whether Combined Transport is good at exchanging high-quality, real-time data. Drawing on his experience from the EU–Ukraine Solidarity Lanes project, Schamschula says the answer is no. Data was often inconsistent, delayed, and manually entered, with even "real-time" updates merely shared hours after events. Crucially, this information rarely reached shippers—the parties who needed it most.

Second, he questions whether operators are moving beyond Excel-based systems. Here too, his answer is no. He observes that rail and intermodal transport still function in isolated digital ‘islands’, with limited integration into broader supply-chain software. While platforms like Railflow are helping expand these islands, he says the sector still lacks true interoperability.

Third, he asks if the industry is ready to replace low- to mid-qualification roles with AI assistants. Although acknowledging the efficiency benefits, Schamschula doubts such transformation would be socially acceptable or supported by unions, giving another firm no.

Summarising, Schamschula concludes that AI is a necessity, not a revolution.

14:55 Aldo Puglisi, Head of IT and Digital at Hupac Intermodal, presents a practical AI prototype developed in collaboration with students from SUPSI University in Lugano. The project focused on creating an AI-powered chatbot to support train monitoring operations by providing real-time access to train data, especially during delays or disruptions.

He explains that Hupac already manages a vast amount of logistics data, from EDI streams and GPS tracking to internal booking systems and partner documents like PDFs and emails. The company’s strategy is to centralise all this information on its digital platform, ensuring both internal users and customers have a unified data source.

To enhance usability, Hupac developed a multilingual chatbot prototype (Italian, English, and German) capable of understanding natural-language queries and retrieving precise, structured answers from internal data sources. The system uses a retrieval-augmented generation (RAG) approach, a hybrid between generative AI and company-specific datasets, allowing users to interact with Hupac’s operational data securely and intuitively. The goal is to make complex data instantly accessible, helping staff react faster to disruptions and improve decision-making within intermodal operations.

15:00 Puglisi continues by outlining the advantages and challenges of using RAG models in intermodal logistics. Among the main benefits, he says, is the ability to interact with company data more intuitively, improving how employees access information and simplify workflows. This, he adds, also represents a shift in staff skills, as teams learn to collaborate with AI tools rather than just use them.

He notes that RAG models allow large language models (LLMs) such as ChatGPT or Gemini to be adapted to a company’s specific domain, increasing their accuracy for intermodal operations. However, key challenges include implementation costs, data quality, and the need for data cleaning and preparation, since noisy or inconsistent data can significantly affect performance. Another limitation, he adds, is the computational load required to run such models efficiently.

For the prototype, Hupac used a representative dataset containing irregularity and disruption records in German, English, and Italian, divided by category. Users could query the chatbot with operational questions such as:

• "Where is the unit located now?"
• "What’s the status of the traffic on this line?"
• "How many dangerous goods are on this train?"

The chatbot could distinguish between historical and real-time data, drawing from emails, PDFs, and internal systems for past incidents, or directly from GPS APIs for live tracking.

15:05 Puglisi shares several examples where the AI chatbot successfully identified causes of delays, such as wagon door issues or vandalism cases, and even confirmed the presence of inspection reports or surveillance footage. The chatbot also demonstrated contextual learning, refining its responses when users repeated or rephrased questions. The team tested the system on multiple LLMs, including Google Gemini 2.5 Pro and several open-source models, finding that while Gemini performed best, some smaller models also yielded promising results within limited datasets.

Puglisi ends his presentation by saying that even small-scale pilots like this demonstrate how AI can enhance situational awareness and efficiency in intermodal operations, provided that data quality and domain expertise remain central.

15:15 Sarah Berger, Intermodal Terminals Specialist at INFORM, introduces her presentation on how artificial intelligence and mathematical optimisation can enhance train loading and network planning in Combined Transport. She presents insights from the KiBa research project, led by Kombiverkehr in cooperation with partners including TU Darmstadt, Goethe University Frankfurt, VTG, and others.

She begins with a familiar scenario for operators: allocating a booked load unit to one or several train trips with the fastest possible routing and a suitable wagon slot. This is far from simple as Combined Transport operates in a highly dynamic, complex environment, where even small changes can ripple across the entire supply chain. Berger outlines the major challenges: multiple actors and routes, volatile demand, and capacity constraints such as train length and yard space. Manual planning, often leads to inefficiencies, safety risks, and human error, especially under disruption or congestion.

To address this, INFORM and its partners combined AI with mathematical optimisation algorithms to assist operators in making faster, smarter decisions. The goal is to use existing infrastructure more efficiently and reduce manual workload in daily operations.

15:20 Focusing on the project’s first area, network planning optimisation, Berger explains that it enables a shift from isolated, manual planning to a data-driven, predictive approach. By distributing load units intelligently across the European rail network, operators can move more containers using existing capacity, reduce congestion and rehandling, and make rail freight faster, more resilient, and cost-effective, even in the face of disruptions.

She explains how INFORM’s optimisation tools work in practice, combining forecasting and optimisation into a continuous, data-driven process. In the first stage, machine learning analyses historical booking data to identify customer behaviour and predict future demand, including shipment weight, type, and origin-destination pairs. These forecasts are then combined with existing bookings, enabling the system to assign load units to train journeys while respecting operational constraints such as train length, weight limits, and delivery deadlines.

She then moves to the second focus area: train loading optimisation, which determines where to place each load unit on a train. The algorithm takes into account booking and wagon data, route profiles, safety rules, and maximum weight limits, automatically generating an optimal load plan. It can instantly recalculate if the wagon configuration differs from what was planned or if a high-priority load arrives late, reducing crane travel distances and rehandling.

Berger adds that this tool also enables centralised train-load planning across multiple terminals, replacing fragmented local operations.

15:30 Osman Akdemir, Co-Founder and Co-CEO of Rail-Flow, opens by saying the transport sector should not fear AI but learn how to guide and apply it strategically. With around 40% of Europe’s rail workforce expected to retire in the coming years, he argues that companies must rely on AI to maintain service quality, efficiency, and 24/7 operations despite labour shortages.

Akdemir introduces the idea of agentic AI,  systems that act like process owners, supporting staff in daily decision-making. He explains that Rail-Flow’s all-in-one intermodal platform already integrates such tools, used by clients like RCG, Hupac, Railion, and E2 Group.

The platform’s AI-driven functions include:

• Autonomous data processing that converts unstructured inputs such as booking updates or wagon lists into structured information;
• Automated integration that triggers follow-up actions like updating train compositions or sharing data with arrival stations;
• Autonomous loading suggestions that assign containers to suitable wagons, considering technical restrictions and safety rules.

Rail-Flow is now testing agentic AI pilots that can recommend workflow improvements, alternative loading strategies, or assess the impact of booking changes. Soon, customers will use chatbots in the portal to manage bookings and estimate time or cost implications. Akdemir agrees with earlier speakers that AI itself is not a competitive advantage but a necessity: Europe’s intermodal sector faces resource shortages and growing complexity, so decisions must be made faster.

15:45 Bart Grauwels, Software Project Team Manager at Camco Technologies, opens his presentation by explaining that automation in intermodal terminals depends equally on software and hardware innovation. Camco, active worldwide with over 300 completed automation projects, designs and produces both hardware and software in-house, reinvesting heavily in research and development. The company’s technologies span OCR and camera portals, real-time location systems (RTLS), and self-service and vehicle booking systems, all forming the foundation for its latest advancement: Automated Damage Inspection (ADI).

He introduces two key software platforms supporting Camco’s automation ecosystem: Bridge, which integrates all automation solutions at a terminal, and the Digital Twin, which allows operators to monitor, replay, and simulate terminal transactions. AI, he notes, has been part of Camco’s DNA since its founding in 1999, starting with feature extraction and image segmentation techniques long before neural networks became mainstream. Early applications included license plate, container ID, and wagon number recognition, later followed by ADR (dangerous goods) label detection.

The company’s first neural network–based solution arrived in 2020, when Camco developed its proof of concept for container ADI (Automatic Damage Inspection). By 2023, all OCR products had been migrated to neural networks, and new camera hardware with GPU acceleration allowed for faster and more complex image analysis.

15:50 Grauwels explains that ADI models are trained using Camco’s own image datasets, ensuring accuracy and reliability. The process involves collecting and labelling thousands of damage samples, training neural networks to recognise damage by type, location, and severity, and then deploying the trained models directly on field cameras, eliminating the need for additional server infrastructure. The same models can run across gate, rail, or crane systems, making them easily scalable.

In practice, the ADI system automatically detects whether a passing container or trailer shows no damage or specific types of damage, forwarding results to the Terminal Operating System (TOS). For significant damage, an operator reviews and confirms the findings, while minor issues are logged automatically. He says that defining what counts as "damage" is a critical challenge. Customers differ in their classification—some use a simple "major/minor" scale, while others apply detailed coding systems such as CDX. Camco solves this with a mapping step between the company’s universal ADI model and each customer’s internal standards.

16:00 According to Grauwels, the system identifies four key damage properties: type, component, location, and severity, which can be mapped to client-specific formats and identifiers. This flexibility allows operators to align the AI’s findings with their own damage assessment procedures.

Grauwels concludes that ADI serves as an assisted, not autonomous, tool, supporting operators by identifying damage rapidly and consistently while maintaining human verification for critical cases. This combination, he says, ensures both efficiency and accountability in automated terminal operations.