Editorial statement to ensure the ethical usage of Large Language Models as part of analytics applications.
Ford Motor Company has developed a virtual digital twin model as an integrated vehicle analysis tool to simulate various driving scenarios of its electric vehicles. To ensure that the digital twin model functions accurately as a twin of the actual vehicle, the simulator output must closely match driving data collected during real-world trips. Achieving this requires tuning of the parameters used within the simulator, a process known as parameter calibration. One unique challenge in this calibration process is the multilevel hierarchical structure, where some parameters must be universally calibrated using data from all trips, whereas others are specific to certain trips. This study devises a new calibration procedure by accounting for this hierarchical structure and demonstrates its superiority using real-world vehicle trip data compared with other alternative methods. Our approach enabled Ford Motor Company to identify key vehicle attributes and understand how trip-specific parameters vary under different driving conditions. With this capability, we also demonstrate the estimation of trip-specific parameters for future trips, which can be used to manage vehicle conditions, such as cabin temperature.
This paper addresses the problem of customer churn faced by Geiger GmbH, a German midsized promotional products distributor. In its drop-shipment business, Geiger was losing 20%-25% of annual turnover, and replacing these customers through acquisition consumed significant sales and marketing resources. To address this challenge, Geiger launched a project to predict and prevent churn, aiming to turn reactive retention efforts into proactive customer management. Our methodology followed the cross-industry standard process for data mining framework and systematically developed a machine learning churn prediction model based on transactional, demographic, and interaction data from Geiger's operational systems. We engineered 77 features that reflect customer recency, frequency, monetary value, and engagement patterns. We benchmarked promising off-the-shelf models, including logistic regression, random forest, XGBoost, LightGBM, and CatBoost, using a time-based, nested cross-validation design. Random forest was selected for deployment because of its robust predictive accuracy and track record in comparable problem settings. The model is integrated into a Microsoft Power BI dashboard, in which it provides sales representatives with automated monthly churn risk scores and actionable watch lists. The results show that the deployed model identifies more than twice as many true churners as random selection when targeting the top 50 customers at risk for churning, maintaining a low false-positive rate. Broader targeting scenarios confirm consistent lifts above 1.8, demonstrating significant operational value. Geiger now proactively manages at-risk customers, reducing unnecessary acquisition costs and increasing salesforce efficiency. The case illustrates how structured machine learning integration can transform retention strategies in noncontractual business-to-business contexts, offering managers a replicable approach to balance churn prediction accuracy with practical deployment considerations.
As the Department of Defense's (DoD) newest branch, the Space Force is being digitally transformed through innovation and data-driven decisions. Space Force Business Analytics teams are enabling and driving digital transformation through up-skilling initiatives, process improvements, and new digital platforms such as dashboards. Although the DoD is not a commercial entity, it faces similar obstacles such as changing the culture, data stovepipes, and platform challenges that are common among transformation participants. After two years of research, implementation, and change, business analytics projects are saving at least 53,000 labor hours each year.
The transportation sector has been the main contributor to emissions growth in the last decade. The type of truck and its delivery characteristics largely explain the transportation carbon dioxide (CO2) emissions and carbon intensity factors. This article introduces a novel methodology for the allocation of a fleet of vehicles to certain regions aimed at minimizing total transportation-related CO2 emissions. Our methodology employs geospatial analysis and machine learning to assess the fuel efficiency and CO2 emissions performance of a vehicle fleet by analyzing historical GPS data, cargo, and fuel use. Subsequently, we include these variables into a mathematical model to obtain an optimal allocation that minimizes total transportation CO2 emissions. Our approach extends the current literature by considering detailed data for operation, such as gradient variability (road hilliness), vehicle speed, elevation/altitude, and distance between stops. We applied our methodology in Coppel, one of the largest retailers in Mexico, which operates its own fleet. Our results showed that, by exchanging 10 vehicles for one month, we observed 8% savings in fuel efficiency and transportation CO2 emissions.
In honor of Gene Woolsey, long-time contributor to Interfaces, I have prepared a "Fifth Column"-style article. This article recalls the many nontechnical hurdles that had to be overcome before developing a decision support model to improve delivery schedules for a small manufacturer. It discusses the importance of organization, process, biases, and key performance indicators on model implementation and acceptance. It makes actionable recommendations on the proper organizational and analytical frameworks from which to view a problem to solve it effectively.
As the COVID-19 pandemic demonstrated, having the right containment resources, such as testing kits, vaccines, and personal protective equipment like masks, at the right time and location to achieve effective pandemic containment and to mitigate pandemic spread is both critical and challenging. High-quality predictions are essential for allocating such scarce resources, and, as our narrative review will demonstrate, substantial work has been conducted on descriptive and predictive analytics during the COVID-19 pandemic. Policymakers attempted to incorporate the resulting information into immediate decision making in reaction to the predicted spread. However, to be prepared for the next pandemic, it is crucial to focus on prescriptive analytics that account for the dynamic interaction between predicted pandemic development and prescribed resource allocation. As we will show, this integrated view has been limited in the literature. To this end, we analyze analytics research conducted during and after the COVID-19 pandemic and its impact on resource allocation. Furthermore, we identify future research streams for resource allocation management, especially those that require close interaction of descriptive and predictive analytics.
Effective pricing and promotion planning constitutes a central pillar of strategic revenue management for firms operating in highly competitive and dynamic markets. These planning activities require the simultaneous consideration of demand elasticity, competitor actions, channel and market specific constraints, and financial objectives. As the dimensionality and interdependencies inherent in these problems increase, manual or traditional approaches become suboptimal and insufficient. In this context, operations research provides a robust methodological foundation for scalable data-driven decision support systems that can optimize complex planning processes across large product and customer portfolios. This paper presents two large-scale optimization systems developed and deployed at PepsiCo to support revenue growth management initiatives: PromoAI and PricingAI. PromoAI integrates machine learning-based promotional forecasts with a mixed-integer linear programming model to optimize promotional calendars across trade channels. The system navigates through millions of product-promotion-timing combinations to find the one that maximizes PepsiCo and retailer revenues, subject to a wide range of customizable business constraints encoded in a modular, user-configurable interface. On the other hand, PricingAI focuses on the optimization of base prices across product portfolios over multiperiod horizons. The system employs Bayesian hierarchical models to estimate own and cross-price elasticities and captures competitive interactions at the product level. These elasticity estimates are fed into a nonlinear programming optimization engine that recommends price changes aligned with revenue and margin targets while incorporating operational constraints such as price thresholds, volume or profit margins, and channel-and market-specific business rules. Together, these systems demonstrate the feasibility and scalability of advanced optimization in large-scale enterprise environments. They highlight the value of integrating statistical learning with mathematical programming to enable enterprise-level automated decision making that is both data informed and aligned with strategic business objectives.
The discipline of operations research and management science (OR/MS) stands at a critical juncture. Although our discipline possesses an unprecedented toolset, sophisticated analytical capabilities, and methodological rigor, a persistent gap separates academic research from organizational practice. This editorial advances a three-part argument. First, the academic-practice gap represents not merely a communication challenge but a fundamental misalignment between how we measure success in academia and the value we could deliver to organizations. Second, this gap stems from identifiable structural factors- incentive systems that prioritize theoretical contribution over practical applicability, limited exposure to organizational contexts, and research trajectories that favor mathematical elegance over implementation feasibility. Third, and most importantly, closing this gap requires coordinated action across multiple levels: faculty must cultivate sustained engagement with practice, doctoral programs must prepare scholars who can bridge theoretical and practical worlds, and institutions must restructure incentives to reward research that matters beyond academic citations. This is not a call to abandon rigor but rather to ensure that rigor serves relevance: that our research fulfills OR/MS's founding promise to help organizations make better decisions through advanced analytical methods.
The present work addresses the automation and optimization of the physician scheduling process in the department of a German hospital using a mixed-integer programming model. We demonstrate how relational data modeling principles can be applied to structure complex scheduling information, enabling the formulation of effective optimization models that would otherwise be impractical. We incorporate this automation and optimization into a decision support tool. Our approach enables the creation of plans much faster with less work, and the accompanying visualizations allow for easier evaluation of plan quality, providing significant managerial insights. This work results in a model that can be swiftly customized and implemented for other hospital-internal departments, as well as divisions in other hospitals. The schedules are built on previous schedules and consider compatibility with shift assignments occurring after the planning period. Furthermore, we introduce the capability to manage fairness over our considered planning horizon.
Selecting the right trials greatly impacts a cancer center's ability to fulfill the mission of helping save the lives of cancer patients, but it is a complicated process involving multiple stakeholders and often lacking systematic analysis. We worked with the clinical and administrative leadership of Hollings Cancer Center (HCC) at the Medical University of South Carolina (MUSC) to revamp the process of trial prioritization and selection to maximize the positive impact of HCC's infrastructure on serving the cancer patient population across South Carolina and the surrounding states. At the core of our project was systematic development and deployment of a multiple-criteria prioritization tool to identify high net impact trials with respect to clinical outcomes and resource usage. HCC has integrated our framework and tool into its trials selection process. It has resulted in more stringent selection of trials (a decrease from a 26% to 17% approval rate), an increased percentage of closed trials meeting/exceeding patient accrual target (an increase from 20% to 39%), and a reduced accrual variance vis-a`-vis target (a decrease from 43% to 29% less than the target).
Just-in-time (JIT) principles are widely used for inventory management, including in healthcare to optimize surgical material purchasing. However, JIT has yet to be applied to the preparation of surgical case carts, containing all the material needed for a specific surgery. These are typically assembled one day in advance, increasing the risk of losing sterility, incomplete carts due to ongoing material use, and corridor congestion. In collaboration with a Belgian General Hospital, we explored the feasibility of JIT case carts, where materials are gathered just before surgery. Our study began with direct observations to identify challenges and opportunities. We then used simulation to evaluate 576 JITimplementation alternatives, varying in buffer type and staffing schedules. Of these, the 217 alternatives that ensured timely readiness in the simulation were benchmarked using a data envelopment analysis with three outputs (buffer efficiency, required corridor space at peak hours, and overall space needed throughout the day) and one input (personnel costs). Seven alternatives emerged on the efficiency frontier, with one ultimately implemented after two pilot studies. Compared with the standard practice previously used, our improvement initiative led to a 67% cost reduction, reduced congestion in corridors, the elimination of incomplete carts, and a decreased risk of contamination by avoiding the stacking of materials. Furthermore, the new method does not require any nurses, so their availability for their core tasks increases. Overall, the JIT approach enhanced surgical logistics and addressed nurse shortages, making it a promising innovation for hospital efficiency.
In 2024, Maryland's capital city Annapolis transitioned two underutilized fixed bus routes serving 133 stops to a microtransit service offering on-demand transportation among 237 stops. Microtransit uses advanced communication and scheduling technologies to dynamically adjust vehicle routes and schedules based on real-time passenger requests. In this paper, we present an mixed integer programming approach to replace the previous insertion heuristic in Annapolis's microtransit operations. Based on one month of operational data, this new approach resulted in a significant increase in the number of trips served, with an average daily rise of 6% in one service area and 19% in the other. Simultaneously, vehicle miles traveled per served trip decreased by an average of 14% in one area and 31% in the other, demonstrating a more efficient use of resources, as the system serves more passengers while reducing the distance traveled per served trip. Moreover, microtransit offers a more personalized service, which tends to attract additional passengers due to its flexibility and convenience. However, as demand increases, the system can become overloaded, resulting in a significant number of trip requests being rejected. In such cases, the city of Annapolis may need to revert from microtransit to fixed bus routes. To support this fallback strategy, we further propose a mathematical programming model based on the orienteering problem to redesign the city's fixed bus routes to best serve observed demand. Our simulation results show that the revised fixed bus routes not only accommodate a larger portion of the demand but also reduce passenger walking distances by 29% for one service area and 24% for the other.
The Master of Science in Business Analytics (MSBA) program at the National University of Singapore (NUS), offered through the NUS Business Analytics Centre (BAC), was awarded the prestigious INFORMS UPS George D. Smith Prize in 2025. At the core of the program's pedagogical philosophy are a structured industry collaboration framework and an "industry-ready pipeline" underpinned by the ACDIN training pedagogy. Together, these represent institutional innovations that effectively bridge the gap between academia and industry, systematically cultivating high-quality, practice-oriented analytics, operations research (OR), and artificial intelligence (AI) professionals. This paper details the program's curriculum design, industry partnership ecosystem, and governance mechanisms, demonstrating how these elements collectively enable the consistent translation of classroom learning into meaningful, real-world impact.
Unsegmented emergency medical services (EMS) data obscure whether certain patients need care more than others, historically limiting cities from pursuing some preventative health interventions. The operations community has long pioneered prescriptive analytics for EMS response using data on incident location and time though novel descriptive analytics that segment demand by patient demographics and exposure type can inform preventative policy efforts, such as making capital investments and conducting outreach to lessen the risk of people being struck by motorists. Segmentation is the analytics response to demand heterogeneity in operations. By linking ambulance records to patient age and neighborhood socioeconomic characteristics, Boston EMS and partner agencies developed a data-driven approach to segment demand and guide siting preventative interventions. The approach parses demand by the demographics of patients struck by motorists (age and neighborhood socioeconomic status) along with the exposure type (walking or cycling during school commute hours) and the location where they were struck. The process overcomes long-standing data handling, computational, and conceptual barriers. Traffic crashes disproportionately affect children walking and cycling to school who reside in Boston's poorest areas. The city used these findings to locate six initial capital projects and identify seven schools in the most affected areas in which to conduct outreach. The city also expanded data sharing between agencies to enable regular preventative work. Compared with the current resource allocation process, this socioeconomically sensitive segmentation that captures the profile of patients struck by motorists elevates disadvantaged neighborhoods in the urban planning process.
This 14th Rothkopf Rankings continues a 30-year tradition that Michael Rothkopf began in 1996 of measuring the contributions of academic institutions to the research on the application of operations research, management science, and analytics. I assess the activity of universities in the production of practice-centric operations research over two different overlapping seven-year periods-2015-2021 and 2018-2024-to maintain consistency with prior Rothkopf Ranking studies. I use the evolving ranking methods, presenting three different measures of output and one blended ranking. Although there are some very familiar and regular top contributors (notably, the Massachusetts Institute of Technology, the Colorado School of Mines, the Georgia Institute of Technology, and the Naval Postgraduate School, among others), there is considerable randomness in the coverage of applied research across most ranked institutions.
Our industry partner seeks an investment strategy and corresponding extraction schedule at daily fidelity of three-dimensional, notional blocks of ore (and waste) to maximize the discounted ounces of metal subject to spatial precedence, geotechnical, and operational constraints. This study develops an integer program, which we implement in Python, that enables fast parametric analysis, supports project-specific constraints, and generates (near-)optimal block schedules. Our model produces reliable and sustainable long-term scheduling solutions within five hours, which is faster than the time required for engineers to manually generate schedules and is acceptable for scenario analyses regarding changes in commodity price, crew productivity, plant sizing, and equipment availability. The broader implication is a 65-fold increase in scenario throughput, scaling from 30 to more than 2,000 evaluations per man-week, allowing for rapid quantification of trade-offs. In a representative operation, the tool showed diminishing returns to scale beyond 2-million-metric-tons-per-year plant capacity and identified a 1.5-million-metric-tons-per-year configuration as the most capital efficient, enabling faster, more confident evaluations of plant size, capital strategy, and operational trade-offs. The success of our optimization model demonstrates the utility of targeted in-house decision support tools for capitalintensive projects, especially when customization, scalability, and cost present challenges.
Pilatus Aircraft Ltd. (Pilatus) is an independent subsidiary of the Pilatus Group, founded in 1996 with headquarters in Broomfield, Colorado. The company markets, sells, and maintains all Pilatus aircraft in North and South America. The company seeks to improve on-time deliveries of all its repair parts, with an emphasis on its high-importance Priority 2 (P2) orders, that is, critical parts associated with an aircraft-on-ground status. A (Q, R) inventory review model, that is, one in which an order of Q is placed when inventory falls below R, improves the mix of parts held at the Broomfield warehouse to more quickly service customer orders; a separate optimization model correspondingly motivates antiquated part liquidation. Over the course of 18 months, our solutions improve on-time deliveries of all order types by 4.7 percentage points while specifically improving on-time deliveries of Priority 2 orders by 5.9 percentage points. Correspondingly, during the most recent year at the time of this writing, that is, 2024, by its own independent calculation, Pilatus reduced the value of goods held in inventory by nearly 21% (excluding the decrease owing to liquidation). The company uses our models as inventory stocking and liquidation guidance.
