Past and ongoing events, both: man-made risks and environmental-caused hazards, have shown the vulnerability of global supply chains and supply chain networks. Since the COVID-19 crisis, the notions of robustness and resilience moved into the center of attention in the management of supply chains and extended enterprise networks. And in fact, nearby events such as global conflicts, false promises, taxes and tariffs, etc., the engineering of robustness-increasing measures and resilient-increasing measures into (global) supply chains are more relevant than ever (i.e., World Economic Forum (2013)).

Both concepts – the concept of organizational robustness and the concept of organizational resilience – are about a system's reliability in withstanding and coping with hazardous disruption. The notion of robustness, in addition, adresses the organization's stability: it is the organizational ability being either insensitive (Gremyr & Hasenkamp, 2011) and/or strong and sturdy to deviations (Christopher & Rutherford, 2004), to withstand risks (Wieland & Wallenburg, 2012), resist disruptions (Vlajic et al., 2012), prevent undesirable impacts (Roy, 2010) as well as retaining the system's structures and funcitons intact (i.e., Meepetchdee & Shah (2007), Vlajic et al. (2012), unchanged, or nearly unchanged (i.e., Wieland & Wallenburg (2013), Klibi et al. (2010)). It is about sustaining operations when a major disruption hits (Scholten et al., 2014), continuing operations, and remaining effective for all plausible futures (i.e., Wieland & Wallenburg (2012)). Under the (realistic) assumption that not all risks, hazards, and events can be prevented, a system is requested to develop routines for organizational responses (Lengnick-Hall & Beck, 2005): to survive, adapt, and grow in unforeseen events, even in catastrophic incidents (Fiksel et al., 2014), thus to be resilient. The notion of resilience is about the system's capability to evolve out of organizational hazards and crises without adverse effects. Colloquially: bouncing back (Sheffi & Rice, 2005) through anticipation and adjustment to trends, capacity to change (Välikangas & Hamel, 2003), self-renewal, adaptations, and continuous innovation (Ponomarov & Holcomb, 2009), (encouraging) entrepreneurial behavior (Reinmoeller & van Baardwijk, 2005) – under certain and uncertain environments (inherent resilience versus adaptive resilience (Ponomarov & Holcomb, 2009)).

Within this paper, we consider the concept of robustness as a resource-based property of an organization: it is about the maintenance of organizational performance, its monitoring, and measurements toward the right product, in the right quantity, to the right price, in the right time, and with the expected quality. In the center are, for example, risk anticipation and mitigation measures, such as organizational risk management and business continuity planning. In contrast, we consider the concept of resilience as a capability-based property of an organization: it represents an organizational governance framework for learning and knowledge, renewal, change, and innovation, orchestrating and mediating between organizational robustness and the organization’s ability to be flexible and agile towards renewal, change, and innovation.

We consider the concept of robustness and resilience as an organizational process of active decision-making.

Therefore, we connect the concepts of robustness and resilience to decision theory – especially to the Theory of Games. Game Theory can be traced to v. Neumann (1928) and v. Neumann & Morgenstern (1944), who established Game Theory as an academic discipline. Game Theory investigates real-life, interactive situations wherein opponents (minimum: two actors) compete for a strategic resource. Under the assumption of rational behavior, the actors have to decide among strategies (minimum: two strategies) to leverage the best, self-interested possible outcome. Related to the game's conditions, the decision-making of the actors can lead to a payoff equilibrium between the players, an asynchronous payoff equilibrium between the players (i.e., one actor is satisfied by the outcome, the other actor is disgruntled), or a dysfunctional payoff equilibrium. Scholars and academics have further developed the theory and introduced important solution concepts to mediate conflicts and cooperation. Solution concepts include, for example, the Minimax Theorem (v. Neumann & Morgenstern, 1944) intoducing safety and security in decision-making; the Nash equilibrium (Nash, 1950a, 1950b) introducing stable cooperation in decision-making; the Shapley value (i.e., Shapley (1953), Hart (1987)) introducing fairness and cooperation in decision-making; the »sure-thing-principle« introducing the concept of strict dominance (Diekmann, 2009).

We turn our focus to rationality in decision-making for the increase of organizational robustness and resilience in times of organizational prosperity and organizational adversity. The leading research question investigates »how managers can decide rationally towards increased organizational robustness and resilience in systems and among ecosystem thus to better respond to and act on organizational dynamics, risks, uncertainties, and crises?«. In responding to the research question, we make use of four business cases. These business cases, elaborated with three Transport Logistics Providers from Austria, Germany, and Switzerland, allow us to connect challenges in rational decision-making with the concepts of organizational robustness and resilience. By transforming the cases into a game, we make use of the Game Theory's 2x2 payoff matrix. In doing so, we include the important aspects of the situation and exclude the unimportant ones (Samuelson, 2016).

At the current state, the paper’s audience is scholars and students within the fields of Supply Chain Management and (Transport) Logistics. After the active integration of the paper at hand into a lecture entitled “Supply Chain Resilience” (5 ECTS course, organized on the 17-19th March 2026 at the Satakunta University of Applied Science (Finland)), its demonstration by making use of the MIT’s Beer Distribution Game and the discussion of the results with staff and students, the paper shall be reworked and submitted to an academic journal. The audience then is both academia (i.e., scholars, scientists, students) and practitioners (i.e., managers and organizational decision-makers) within the field.

Acknowledgements:

This work is supported by the Interreg Alpine Space Project: Apollo – territoriAl corPOrate weLfare through digitaLization and cOoperation. The authors wish to thank the funding authority and people involved in the project and paper for their support and collaboration in making this research possible.

Purpose of this Paper

Efficientcargoloadingremainsacornerstoneofgloballogistics,yetitpersistsasacriticalop-erationalbottleneckduetotheinherentcomputationalcomplexityoftheThree-Dimensional Bin Packing Problem (3D-BPP). Operationally difficult loading decisions cascade into higher transportation costs, wasted container space, and delivery delays. Although recent ad-vancements have produced promising strategies using deep reinforcement learning[6] and heuristic methods[5], these solutions frequently operate mostly as black boxes. They often lack the flexibility to accommodate dynamic, real-world constraints or the implicit knowl-edge held by on-ground personnel. Consequently, the industry often relies on manual planning, which is inherently unscalable and error-prone. The purpose of this paper is to in-troduce FraightMind, an end-to-end intelligent decision support system designed to bridge the gap between algorithmic precision and human expertise. By integrating a conversa-tional Large Language Model(LLM) interface, users can either select standard truck and cargo types from a pre-defined database or simply describe custom dimensions and quan-tities in a natural, conversational manner. Unlike static solvers, it provides an interactive 3D environment that empowers users to intuitively perform real-time what-if analysis or modify the solutions provided by the algorithm to make them operationally better based on their on-ground knowledge. By capturing and storing these user-driven refinements, FraightMind aims to create a scalable, user-friendly architecture that transforms loading process into a collaborative, seamless process between artificial intelligence and human decision-makers.

Design/methodology/approach

The FraightMind framework is architected as a modular, end-to-end decision support sys-tem that unifies Generative AI, a packing algorithm, and interactive 3D visualization. As illustrated in Figure 1, the workflow begins at the Interface Layer, which is a conversa-tional chatbot powered by Llama 3[3] and orchestrated via LangChain[2] serving as the primary entry point. Here, the user provides input through two distinct methods: select-ing from a pre-populated database of standard fleet and cargo types, or conversationally specifying unique dimensions and quantities. This input is processed by the Intelligence Layer, where an LLM-based parser (Llama 3) converts natural language into structured, schema-compliant JSON data.

Once validated, this data drives the Algorithm Layer, where a constraint-aware constructive heuristic generates a near-optimal 3D packing plan prioritizing volume utilization and stabil-ity as depicted in Figure 2. The process sorts cargo by volume and utilizes a layer-building strategy to place items, dynamically managing a heightmap to ensure structural integrity. A local search mechanism resolves placement failures or make the placements better by re-orienting subsets of boxes. The resulting load plan is rendered in the Unity engine[4], creating a physics-enabled environment where users can perform real-time what-if analy-sis. This interactive layer allows for drag-and-drop adjustments with immediate validation against physical constraints (e.g., overlaps, overhangs). The system closes the feedback loop by capturing these manual refinements in the database for future packing strategies. A demonstration video of the complete system workflow is available at [Link].

Findings

The development and evaluation of FraightMind yielded critical findings regarding AI and human interaction in logistics. We observed that decoupling the LLM from core algorithm using it strictly as a semantic translator to extract structured JSON from natural language effectively eliminated hallucinations, ensuring the heuristic solver received only valid con-straints. Chatbot interface helps the user to interact in a conversational flow with the algo-rithm and the solutions. Performance analysis of the packing algorithm demonstrated that the layer-building strategy combined with local search backtracking achieves a nice balance of computational speed and packing arrangement required for real-time web interaction. The integration of interactive Unity-based visualization enabled what-if analysis capability allowing users to inject tacit operational knowledge, which pure algorithms often miss. We found that real-time constraint validation during manual drag-and-drop adjustments sig-nificantly reduced potential loading errors by immediately flagging physical violations like overhangs or instability. Our modular design enables that the packing algorithm can be replaced with future learning based algorithms to take advantage of the human based re-fined solutions. Ultimately, the results confirm that a collaborative human-AI architecture,

2

where the algorithm handles combinatorial complexity while the human operator manages strategic exceptions, provides a more robust, transparent, and scalable solution than fully automated black box solvers.

Value

Our work presents a holistic framework merging Large Language Models with 3D packing algorithms and physics-based visualization. The novelty lies in FraightMind’s architecture, which democratizes interaction with platform based solution for non-technical staff via nat-ural language. The value extends to operations by transforming logistics processes into a collaborative dialogue. By supporting real-time what-if analysis through physics-aware 3D interaction, the system empowers users to visually verify algorithmic solutions. Addi-tionally, the architecture captures human feedback to create a dataset of practical loading patterns, offering immense value to industry practitioners seeking deployable tools and for researchers exploring human-AI collaboration.

Research limitations/implications

The current iteration of FraightMind relies on a constructive heuristic approach for packing. While this ensures computational speed suitable for real-time web interaction,it does not evolve with data especially the one that we are collecting from the human feedback. Future research implications involve integrating learning-based components capable of adapting placement strategies by leveraging the human feedback loop established in this work. Ad-ditionally, there is significant potential to expand the system’s scope by integrating many other allied logistics operations, thereby creating a comprehensive end-to-end logistics so-lution.

Practical implications

FraightMind offers a pathway to modernize logistics by deploying AI as a collaborative as-sistant, significantly reducing planning time and loading errors while boosting container utilization. Additionally, the system facilitates better cross-functional communication be-tween planning and warehouse teams through standardized, visual digital reporting.

REFERENCES

[1] Apache CouchDB. 2025. https://couchdb.apache.org/. Ac- cessed: 09/02/2026.

[2] LangChain. 2025. https://www.langchain.com/. Accessed: 09/02/2026.

[3] Ollama Team. 2024. Ollama: Run LLMs Locally. https: //ollama.com. Accessed: 09/02/2026.

[4] Unity Technologies. 2023. Unity. Game development platform.

[5] Zhu, Q. et al. 2021. Learning to pack: A data-driven tree search algorithm for large-scale 3d bin packing problem. In Proc. CIKM, 4393–4402.

[6] Hu, H. et al. 2017. Solving a new 3d bin packing problem with deep reinforcement learning method. arXiv preprint arXiv:1708.05930.

Purpose – The agri-food sector is experiencing mounting pressure to transform conventional supply chains (SCs) into regenerative and resilient systems. This pressure is driven by escalating environmental degradation, including soil depletion and biodiversity loss, growing social challenges such as unequal power relations and declining farm viability, as well as increasing economic and geopolitical uncertainty (Islam and Zheng, 2025). Regenerative SC concepts go beyond conventional sustainability approaches by not only aiming to reduce negative impacts but by actively restoring ecological and social systems. Such concepts emphasise long-term collaboration, trust and shared value creation across SC actors, requiring deeper relational engagement and long-term exchange. However, implementing regenerative practices often involves higher upfront investments, longer time horizons and increased interdependence, which fundamentally alters the nature of buyer-supplier relationships. Despite the shift towards a more relational approach, there is still limited understanding of how particular relational dynamics can promote the development of regenerative and resilient agri-food supply chains (AFSCs) (Meyer et al., 2025). In particular, supplier loyalty, defined as “long-term supplier commitment”, remains insufficiently understood. While existing SC research (e.g., Ahmed et al., 2024) has extensively examined transactional efficiency, contractual governance and sustainability performance outcomes, relational foundations such as loyalty are often treated implicitly, reduced to trust or long-term orientation or overlooked as “soft” factors. As a result, little is known about how loyalty-based relationships shape stability, risk-sharing and collaborative learning processes that are critical for regenerative transformation. Therefore, the purpose of this study is to examine the role of supplier loyalty in building regenerative and resilient AFSCs. Specifically, the study seeks to uncover how loyalty-based supplier relationships enable the implementation of regenerative practices and resilience across different stages of the AFSC.

Design/methodology/approach – This study uses a qualitative research approach involving interviews with experts from three distinct groups in AFSCs: (a) organisations with indirect influence, such as consulting and advisory firms; (b) organisations with direct influence, including seed producers, farmers, manufacturers, wholesalers and retailers; and (c) organisations in special coordinating positions, particularly agricultural cooperatives. Such a differentiated selection reflects the heterogeneous roles, power structures and interdependencies that characterise contemporary AFSC networks. A total of 17 expert interviews were conducted between January and March 2023 in Germany. Based on their strategic position and decision-making responsibility, interviewees included department heads and managing directors with professional experience ranging from four to 35 years. With a duration ranging from 20 to 70 minutes, interviews were conducted using a mix of digital communication platforms, telephone interviews and in-person meetings, reflecting the post-pandemic transformation of professional interaction. An open-ended approach to interviewing was employed to allow respondents to articulate their perspectives freely while maintaining analytical comparability. To analyse the qualitative data, the GABEK® methodology was employed, supported by the WinRelan® software. GABEK® enables the systematic analysis of complex qualitative data by focusing on semantic meaning and contextual associations (Zelger, 2000).

Findings – The relationship between supplier loyalty and SC resilience emerged consistently across interviews and was described as particularly salient under conditions of heightened uncertainty. In this context, supplier loyalty was not merely portrayed as a relational outcome, but rather as a critical cognitive and informational resource for regenerative supply chain intelligence. As experts emphasised, long-term relationships facilitate the accumulation of tacit knowledge regarding production conditions, risk factors and partner capabilities, which in turn enhances supply chain actors’ ability to anticipate disruptions and align strategic decisions with regenerative objectives. According to experts, loyal suppliers are therefore more willing to adjust production volumes, delivery schedules and quality specifications in response to disruptions. These adaptive behaviours were most evident during periods of extreme uncertainty, such as climate-related production shocks, logistical bottlenecks and regulatory changes. In this context, supplier loyalty was found to mitigate the adverse effects of SC disruptions by facilitating cooperative risk-sharing and informal coordination mechanisms. Rather than relying solely on contractual enforcement, loyal partners were described as engaging in problem-solving behaviours based on mutual understanding and shared interests. The findings further indicate that supplier loyalty is linked to a shift from transactional to relational governance structures. Loyalty-based relationships were highlighted as reducing opportunistic behaviour and fostering a governance environment in which regenerative goals can be jointly pursued. This pattern is particularly relevant given the prevalence of power asymmetries and price pressures that undermine sustainability initiatives in AFSCs. Relatedly, the data reveal that supplier loyalty is associated with a reduced dependence on formal control mechanisms. According to the experts, loyal relationships allow for greater reliance on informal coordination, trust-based agreements and shared norms, which in turn lower coordination costs and enhance responsiveness.

Value – By offering empirical insight into the relational foundations of regenerative and resilient AFSCs, this study provides valuable contributions to both theoretical and practical domains. Moreover, the research transcends the conceptual discourse on regeneration, elucidating the operational dynamics of supplier loyalty as a pivotal facilitator of long-term resilience, ethical governance and adaptive capacity. Building upon extant SC management literature, the study offers an empirical linkage between supplier loyalty and regenerative outcomes, whilst also unveiling qualitative associations between loyalty, resilience and intelligence-driven decision-making. Furthermore, the results of the study demonstrate how loyalty-based relationships can reduce vulnerability to disruptions, support regenerative investments, and enhance collaborative problem-solving across supply networks.

References

Ahmed, A.A.A., Kumar, V.S., Jena, S.K., Nagpal, A., Shukla, P.K. and Balachandar, K., (2024), “Maximizing Profits and Efficiency: The Intersection of AI, Machine Learning, and Supply Chain Financial Management”, Utilization of AI Technology in Supply Chain Management, pp. 225-239.

Meyer, C., Luiz, J.M., Grutter, A. and Parker, H., (2025), “Disintermediation and Reintermediation of Seafood Supply Chains for Social and Ecological Regeneration”, Journal of Supply Chain Management, Vol. 0, pp. 1–17

Islam, M. Z. and Zheng, L., (2025), “Why is it necessary to integrate circular economy practices for agri‐food sustainability from a global perspective?”, Sustainable Development, Vol. 33 No. 1, pp. 600-620.

Zelger, J. (2000). Twelve steps of GABEK WinRelan. In: Buber, R., Zelger, J. (Eds.): GABEK II. Zur Qualitativen Forschung. On Qualitative Research, Studienverlag. Innsbruck, Wien, pp. 205-220.