Abstract Prior research highlights tolerance for failure (TFF) at individual or team level, yet how firm‐level TFF shapes exploratory versus exploitative innovation remains underdeveloped. We conceptualize TFF as a stable organizational cultural norm and operationalize it via a firm's governance responses to executive underperformance, specifically, the absence of CEO dismissal or pay cuts following poor financial results. Drawing on organizational control theory, we argue that TFF influences innovation trajectories through intertwined control dynamics and resource allocation. We further examine the moderating roles of two external control forces: qualified foreign institutional investors (QFIIs) and government subsidies. Using a novel firm‐level TFF measure, we find that TFF promotes exploratory innovation while constraining exploitative innovation. Government subsidies mitigate TFF's detrimental effect on exploitation but do not moderate the TFF‐exploration relationship. For firms with heavy QFII holdings, QFIIs attenuate TFF's positive effect on exploration and weaken TFF's negative effect on exploitation. This study advances innovation literature by integrating a refined TFF construct with organizational control theory, clarifying the contingent impacts of failure tolerance on innovation and offering actionable insights for balancing risk‐taking and accountability in R&D management.
Abstract Digital twins are essential for any firm on its digital transformation journey. This article introduces the concept of digital twins to the readers of the Decision Sciences journal and provides a review of their benefits, foundational elements, and enabling technologies. We examine the literature on digital twin applications, including their use in manufacturing and supply chain management. The article concludes with a discussion of the challenges involved in implementing digital twins and identifies future research opportunities. We expect this article to inspire the Decision Sciences community to engage with and contribute to the growing body of research on digital twins and their applications.
Abstract The world generates substantial waste that could serve as feedstock, but often inefficiencies in waste exchange arise, caused by a lack of coordination among companies. When trading waste reuse, negotiators must manage the trade‐offs between supply chain, personal, and environmental outcomes. Unlike studies focusing on monetary incentives (e.g., taxation, subsidies), we aim to identify how nudge techniques impact negotiation agreements in waste supply chains. We specifically study the effect of various (negative environmental and cost‐adding economic) frames and (collective‐level and individual‐level) social norms on agreement rates. We analyze supply chain coordination in a set of stylized bargaining experiments characterized by moderately conflicting environmental and economic interests. We design different treatments in two studies, varying the environmental and economic frames in Study 1, and the presence of social norms (derived from Study 1's decisions) in Study 2. We find that framing environmental externalities using a damage frame results in higher agreement rates compared to a help frame. However, a low individual‐level norm indicating low acceptance of unfair offers tends to reduce agreement rates. We reveal that framing and social norms affect collaboration by triggering both proposers and receivers' loss aversion in both economic and environmental dimensions. Additionally, we have discovered the effective boundary conditions for the effectiveness of the nudge effects, as well as the impact of decision‐makers' internal motivations (environmental concern, social value orientation) on the coordination decisions. Our findings can guide policy makers to improve inter‐firm collaboration in the presence of negative environmental externalities by using frames to nudge supply chain partners.
Computer-mediated communication channels, such as text-based chat, can improve response speed and scalability while lowering operational costs for customer support. Yet, there has been a growing concern regarding the remote nature of digital communication leading to customer dissatisfaction. This study investigates how communication style-specifically, multiline messaging-can alleviate this concern and affect quality in digital service interactions. Leveraging a unique data set of text-based conversations from a service center, we show that breaking down lengthy statements into multiple lines (i.e., multiline messaging) can improve problem resolution. However, this impact exhibits an inverted U-shaped relationship; although moderate segmentation aids task-oriented problem resolution, excessive multiline messaging backfires by inducing cognitive overload. In contrast, we find that customer satisfaction increases monotonically with multiline messaging intensity. Finally, we demonstrate that multiline messaging interacts positively with agent sentiment to further enhance both problem resolution and customer satisfaction. Our work contributes to the digital service literature by identifying a low cost, communication-based mechanism that can improve service performance, thereby addressing the cost-benefit trade-off often debated in the field. It also enriches the emerging people-centric operations literature by introducing a novel, effective communication construct to the set of human-related drivers of work performance.
This article investigates how a bus firm can best operate its electric bus (EB) charging stations to improve efficiency while serving dual users: EBs and private electric vehicles (PEVs). We model each station as a stochastic service system and formulate a bi-level optimization model to capture the interactions between the bus firm and the PEVs. A multiple-population genetic algorithm is proposed to solve the model. We find that: (a) our sharing strategy can promote station efficiency and ease PEV users' charging anxiety, achieving over 80% utilization of charging resources while ensuring PEV users wait no longer than 0.5 h; (b) reducing the maximum EB service time increases the number of open stations but raises costs and decreases charging stations' profit. Reducing the maximum waiting time for PEVs alleviates the congestion of stations at the expense of revenue. (c) Compared to a benchmark model where a strategy of fixed opening hours and number of shared chargers is used, our model's dynamic optimization of opening hours and charger allocation per time period improves operational efficiency (especially when the transition cost is not high). The proposed algorithm performs better than the traditional genetic algorithm in 50 experiments. Our proposed methods can provide decision support for the operation of shared EB charging stations and promote the development of a green and low-carbon transportation system.
Decision science is entering a new era in which decisions are no longer made solely by humans, but increasingly by autonomous AI (Artificial Intelligence) agents and human-agent collectives. Prior research often treated AI as a tool for prediction or support. Agentic systems now decide, act, learn, and coordinate over time, changing the unit of analysis for decision science. In this perspective, we argue that this shift requires new conceptual foundations and a sharper research agenda. We outline a design-oriented framework that decomposes human-agent systems into atomic structures, decision architectures, and field-to-model mappings, enabling comparison, experimentation, and cumulative knowledge building. Building on this framework, we introduce the mission of the new "Agentic AI and Human-Agent Collaboration in Business" department. The department emphasizes two central topics: agents as decision-makers in single- and multi-agent systems, and human-agent collaboration. We also highlight preferred methods, including lab, field, and simulated experiments, as well as agent-driven research that uses AI agents for exploration, hypothesis generation, and knowledge discovery. The goal is to guide rigorous research on how agentic systems reshape decision-making in business and society.
On quantity-based surplus food platforms, retailers sell surplus bags containing the end-of-day leftovers from their stores. Each retailer's decisions on bag price, production quantity, and quantity reserved for surplus sales are complicated by: (i) interlink between reservation for surplus sales and ability to serve in-store sales, (ii) uncertainty in the leftover quantity while having to ensure a guaranteed minimum quantity for the bag, (iii) intricate relationships among different contextual factors (e.g., consumer consumption needs, consumer valuation of the bag, and platform discount rate), and (iv) competition among retailers. We identify conditions under which the retailer should increase the bag price to improve the profit margin on surplus bags and/or reduce unmet demand from in-store sales, highlighting the need to carefully balance retailer-, consumer-, and platform-specific factors. Interestingly, compared with a no-competition benchmark, competition can lead to a higher bag price, more food reserved for surplus sales, and less food available for in-store sales. More importantly, we offer three key insights regarding resulting outcomes. First, competition (versus no-competition) can yield higher consumer surplus, higher retailer's profit, and lower food waste (e.g., when the discount rate is low, and consumers' valuation of surplus bags is high). Second, intense competition can reduce retailer's profit and exacerbate food waste, suggesting that having more retailers may result in unintended consequences. Third, platforms can mitigate these consequences by leveraging the discount rate. For instance, in areas with many competing retailers and many consumers with small consumption needs, platforms should set the discount rate low.
Synchronization in complex networks is an impactful phenomena with applications in numerous disciplines that have attracted the researcher’s attention for decades. Recently, there have been efforts to investigate brain synchronization by representing brain signals as complex networks and analyzing the dynamic interactions between neural regions to uncover patterns of connectivity and coordination. Focusing on functional Near-Infrared Spectroscopy signals converted to visibility networks, this paper incorporates reinforcement learning into the exploration of synchronization suppression in the visibility networks constructed, enhancing interventions to mitigate excessive neural synchronization. In this study, we extend the Kuramoto model by adding input signals to pinned nodes of the networks generated via the Reinforced Learning, more precisely the Proximal Policy Optimization algorithm, and analyze the synchronization suppression conditions. Comparison to results from other existing models in the literature is done via an experimental study in a realistic setting.
In scientific conferences, participants are often required to select among parallel presentations based on their interests and logistical constraints, facing challenges related to time management, distances between session rooms, and content overlap. Building a daily schedule that maximizes individual satisfaction thus represents a complex and time-consuming task. In this work, we propose a system based on artificial intelligence and mathematical optimization to assist users in creating an optimal daily program. To this end, we develop a hybrid framework that combines explicit evaluations provided by participants with an automatic estimation of interest indices for talks, leveraging Large Language Models (LLMs). Finally, we present a mobile application, which implements the scheduling optimization process and collects qualitative feedback directly from participants, validating the effectiveness of our approach through experiments conducted on real-world conference data.
The rapid growth of online shopping and logistics have significantly raised the question of how to adequately meet customer delivery expectations. Considering the advancements made in drone technology, this challenge could be addressed more efficiently than using traditional vehicle-based delivery technologies. According to analysts, the use of aerial drones can be beneficial as they are capable of lowering the cost of transportation, avoiding stagnation on roads by flying over traffic congestion, and being non-polluting to nature by not consuming any fossil fuel. However, drone delivery optimization models are distinguished for their complexity, due to the nonlinear nature of the energy constraints. In this study, we analyze the computational complexity of the optimization model from [1, 2] and focus on the linearization of the energy consumption constraints, which are a nonlinear function of the payload and the travel time. We discuss and analyze various linearization techniques to reduce the resolution time especially for large data size problem instances.
Society 5.0 represents a vision where technology and human decision-making work together to address complex social challenges. This research examines how Digital Twins can function as effective decision-support tools in this emerging societal framework, using Urban Mobility as an illustrative case. The principles presented for Urban Mobility Digital Twins (UMDTs) can be applied to many complex systems. The digital twin approach extends beyond specific domains—whether managing water resources, optimizing energy distribution, or planning healthcare services—by creating virtual counterparts that learn continuously from real-world data. Digital twins transform how cities approach mobility planning by converting data into practical insights. Planners can visualize potential changes, test different scenarios, and understand impacts on both technical performance and community experiences before implementing solutions in the physical world. This methodology aligns with Society 5.0’s central idea: using technology to enhance human wellbeing. By developing accessible, validated models that connect technical and social considerations, digital twins help create shared understanding among diverse stakeholders and support more informed decision-making.
Objective: This study investigates how stress affects airline pilots’ performance and explores the potential of psychophysiological monitoring tools to assess and mitigate related risk. Background: Modern commercial aviation is a hypercomplex socio-technical system in which safety margins are narrow and operational demands constantly evolve. While technological advancements have enhanced efficiency, they have simultaneously increased pilots’ cognitive load by requiring the rapid integration of vast amounts of information. Stress, intensified by the complexities of the operational environment, undermines pilots’ cognitive performance and constitutes a critical risk to the safety and efficiency of airline operations. Methods: A structured and reasoned literature review was conducted using a snowballing technique to identify empirical and theoretical studies on psychophysiological indicators of stress and cognitive load in aviation. Results: The review synthesises a range of biometric measures applied over time to capture pilots’ cognitive and emotional states. These approaches demonstrate the feasibility of monitoring airline pilots’ stress status in real time and highlight persistent challenges related to intrusiveness, ecological validity, and operational integration. Conclusions: In an increasingly automated and data-intensive aviation environment, distress-related cognitive impairment constitutes a critical operational impediment for safety and efficiency. Addressing this challenge requires human-centred monitoring systems capable of detecting early physiological distress markers to enable adaptive interventions. Biomarker-based multimodal methods show promise, but their effectiveness relies on trustworthy data integration and accurate contextual interpretation. Emerging approaches, such as Deep Metabolic-processes Assessment (DMA®), may advance non-invasive detection of stress-induced metabolic changes. Recognising pilots as central agents of system resilience underscores the need to integrate cognitive state monitoring into future aviation system design to enhance safety and efficiency.
Optimal management of large warehouses or regional distribution centers requires to optimally solve a number of interrelated sub-problems such as deciding the layout of the warehouse, i.e. the organization of aisles and shelves, the number of items to be made available for each stock keeping unit (SKU), the location of each SKU in the warehouse and the routing policies to be followed by pickers. These sub-problems are usually analyzed separately in the scientific literature. Moreover, real cases are often characterized by additional constraints and complicating features that are seldom considered in academic papers. We present an attempt to optimize them jointly by means of suitable mathematical optimization models and solvers. The method has been used to optimize the operations in a warehouse with an irregular layout and precedence constraints between different materials. Relevant improvements in the productivity of pickers were observed when the suggestions stemming from this analysis and optimization were compared with the company current practice.
As artificial intelligence tools generate increasingly complex risk scenarios, project managers face a critical challenge: hundreds of potential risks per project that far exceed human capacity to evaluate manually. While other high-risk domains have developed frameworks for human-AI collaboration in high-volume contexts, project management lacks systematic approaches to this emerging problem. Without proper frameworks, AI adoption risks creating an “intelligence paradox”—where AI’s analytical power generates so many scenarios that managers become more overwhelmed than before. This study presents a cross-domain framework transfer analysis through a systematic review of 344 papers. We identify four proven human-AI collaboration frameworks from cybersecurity, robotics, security screening, and financial systems that successfully manage high-volume risk data. Our analysis reveals that while 17 studies document information overload in project management, none provide systematic solutions adapted from other domains. To address this gap, we develop a comprehensive framework consisting of: (a) a transfer matrix that maps each source framework to specific project management challenges, (b) an integrated system architecture preventing AI-induced overload, and (c) three critical principles—context preservation, threshold calibration, and integration requirements—that guide successful implementation. The framework provides practitioners with concrete pathways for adaptation: from cybersecurity’s HAT framework for risk prioritization to weak-signal detection for early warning systems. This study offers the first systematic approach to transferring proven human-AI collaboration models to project management, providing both immediate solutions and a methodology that other fields facing AI-integration challenges could adapt.
Parametric insurance is attracting interest as a device to expand coverage for catastrophic events. These policies pay based on a few measurable characteristics (parameters), rather than on the loss incurred. Since these characteristics are usually obtained quickly through a public and reputable technical agency, parametric mechanisms offer more transparency and speed than traditional insurance products. However, these positive features come at the cost of precision. Since there is no adjustment of the policy in the classical sense, the payout produced is commonly calculated using algorithms, models, or equations that aim to approximate the loss suffered by the insured. The insurance industry refers to this potential inaccuracy in recoveries as “basis risk,” intuitively understood as the difference between the payout that one would expect from a traditional indemnity policy and the payout actually received from the parametric coverage. In this paper, we offer a formal description of basis risk using the Loss Capture Ratio (LCR), a metric that we have used successfully at Guy Carpenter to illustrate the performance of the parametric solutions we design for our clients.
Currently, the development of intelligent production systems based on modern artificial intelligence methods is an urgent problem. Industrial automation has its own specific limitations in the application of artificial intelligence at the level of software and hardware implementation. The creation of a new modified endocrine-immune algorithm (EAIS) for complex objects control and diagnosing the state of equipment is a promising and relevant task. The proposed EAIS algorithm is adapted for work with industrial controllers of the Modicon series (Schneider Electric) of the IEC 61131–3 standard. The results of modeling and experiments were carried out in the Industrial Automation Lab (KBTU JSC). The developed algorithm was tested on the engineering database of the Tengizchevroil oil refinery and modeled using the Modicon M340 controller. The advantages of EAIS include the ability to be placed directly in the controller without using external servers for training. For this purpose, a user-defined functional block DFB (Derived Functional Block) is created with a modified endocrine-immune algorithm EAIS_FBD, implemented in the ST (Structured Text) language. The organization of an intellectual structural program unit (Intellectual Program Organization Unit) is proposed for placing the functional block EAIS_FBD and multiple use in the development of process control systems. Comparative analysis of the results of modeling with the support vector method, decision trees and nearest neighbors showed the superiority of EAIS.
In a world where e-commerce and social engagement are increasingly widespread among potential customers, search engines and social networks constitute powerful vehicles for companies’ advertising campaigns. Typically, an advertiser is required to indicate a bid for one or multiple keywords associated with a target ad. Then, a sponsored auction, conducted by the advertising platform and involving multiple advertisers, determines the visibility of each ad, which in turn influences the cost paid by the advertisers for the generated engagement. Designing optimal online bidding strategies, subject to time and budget constraints, can drastically improve companies’ profits. We review scientific papers on this topic from the last two decades, discuss the state-of-the-art frameworks used to approach the underlying optimization problems, and identify key research gaps and opportunities for future investigation.
Parametric solutions emerged in the 1990s as an alternative to the traditional insurance solutions. Parametric solutions offer a faster and more efficient risk transfer approach, not relying on lengthy claims process, but on pre-agreed payouts triggered by specific parameters. This paper explores a regression-based methodology to support the design of cat-in-a-grid tropical cyclone parametric solutions, considering different tropical cyclone characteristics, typically publicly available in the aftermath of the event. The paper specifically investigates different single-metric parametric solutions for Jamaica based on maximum wind speed, minimum central pressure, and radius of maximum wind speed, respectively. However, the methodology is general and can be applied to different regions and hazards. Results show that parametric solutions based on maximum wind speed and minimum central pressure provide good accuracy metrics and a satisfactory agreement with loss estimates for historical events, whereas parametric solutions based on the radius of maximum wind speed have limited predictive power, suggesting a better use of such parameter as a complement of maximum wind speed and/or minimum central pressure in multi-metric parametric solutions.
This paper presents a comprehensive mathematical framework for enhancing machine learning robustness against adversarial attacks through novel theoretical constructs. We introduce Archive Function Robustness theory, which provides formal bounds on the impact of data corruption in learning systems, and develop an extended Dataset Core methodology for efficient processing of large-scale datasets while preserving information integrity. Our theoretical contributions include: (1) a rigorous characterization of archive function stability under adversarial perturbations with provable Lipschitz-based bounds, (2) stratified and adaptive Dataset Core algorithms that maintain ε -approximation guarantees for big data scenarios, and (3) consistency-based verification techniques for poison detection in streaming environments. The framework demonstrates that well-behaved archive functions exhibit bounded degradation under data corruption, with explicit relationships between poisoning rates, data distortion, and system performance. Our approach enables scalable robust learning with theoretical guarantees, providing a foundation for trustworthy AI deployment in adversarial settings.
Metabolic pathways regulate essential biochemical processes in biological systems, and their modeling has provided fundamental tools for understanding cellular metabolism and its applications in fields such as precision medicine, biotechnology, and metabolic disease diagnostics. Over the years, numerous mathematical and computational models have been developed to describe these networks, employing approaches ranging from flux balance analysis to dynamic systems and data-driven methods. This work presents an overview of existing models for describing metabolic pathways, emphasizing their role in representing biochemical processes and their scientific and industrial applications. Building on this review, new perspectives are explored regarding the integration of metabolic models with innovative experimental data, particularly physiological signals that may contain implicit information yet to be validated. One specific case involves the use of surface electrical currents measured on human skin, which could provide insights into the metabolic demand of an organism and require validation through existing modeling frameworks. Furthermore, the potential for integrating metabolic models with advanced computational approaches is analyzed, with a focus on their contribution to the development of Human Digital Twins. These models enable “what-if" simulations to predict the effects of metabolic variations in a controlled virtual environment. This perspective opens new opportunities for research and applications in metabolic modeling, with implications for diagnostics, personalized therapies, and the development of new biomedical analysis tools.
