Making Operations Research More Accessible: Insights from the Rise of Machine Learning

1. Introduction

Operations research (OR) and machine learning (ML) both employ predictive models, have broad applicability, and deliver significant economic value by enabling tasks to be performed more efficiently and cost-effectively than by humans. However, OR remains less accessible to many scholars and practitioners, and its communities, though strong, have not seen the same recent growth and broad engagement as ML communities. OR lacks visibility relative to its real-world contributions, such as on public policy, educational curricula, and research funding. This paper examines several factors contributing to this disparity and proposes a set of recommendations to modernize OR by drawing lessons from the success and popularity of ML, aiming to make OR more accessible.

OR is a quantitative discipline that has been defined as “the scientific process of transforming data into insights to make better decisions” (INFORMS 2024a). Its popularity in academia and practice has grown in the last half century because of the broad range of applications and widespread availability of large data sets (Camm and Watson 2023). The number of employees working in OR in the United States has steadily increased in recent decades, and the demand for operations research analysts is projected to grow by 23% from 2023 to 2033, significantly outpacing the average growth rate for all occupations (Bureau of Labor Statistics 2025b). However, in terms of its size and influence, OR’s growth has been modest compared with related fields.

In contrast, ML has surged in popularity in recent years, leading to widespread adoption in the industry as well as new courses and degree programs at universities. The demand for ML expertise is expected to continue to grow in the United States, with employment in ML projected to grow by 73,100 jobs from 2023 to 2033—more than 2.5 times the increase expected in OR employment over the same period (Bureau of Labor Statistics 2025a, The White House 2025). Several factors explain ML’s rise: the widespread availability of data, breakthroughs in deep learning and neural networks, and the availability of open-source tools and libraries such as TensorFlow, PyTorch, and scikit-learn. These tools have lowered barriers to entry, enabling both developers and researchers to implement, experiment with, and teach ML techniques. ML courses have become foundational to data science degree programs, which have seen a sharp increase in enrollment and number of graduates in recent years (Pierson 2023). Academic interest has followed suit. According to Scopus data, publications referencing “machine learning” as a keyword have grown dramatically over the past decade. In 2024 alone, more than 140,000 ML-tagged papers were published—compared with just a few thousand referencing “operations research” and similar keywords (Scopus 2025). This reflects not only ML’s explosive growth but also its broader adoption across disciplines. Federal research investment mirrors this trend: in the U.S. federal year 2025, AI R&D funding reached $1.95 billion, dwarfing the funding allocated to OR (Holohan 2025).

The unquestionable ascendancy of ML may act either as a catalyst or a disruptor for OR, opening the door to new synergies or, conversely, threatening to overshadow its contributions, marginalize the field, and disrupt its talent pipelines. These risks should serve as a timely warning for OR to strategically showcase its distinct strengths and expand its influence within the evolving data science landscape. By proactively enhancing its accessibility, modernizing its outreach, and highlighting its powerful decision-making capabilities, OR can attract a new generation of talent and forge broader collaborations to instigate important scientific advances.

1.1. OR Definition and History

Historically, OR emerged during World War II, initially defined by the British Army in 1939 as “a scientific method of providing executive departments with a quantitative basis for decisions regarding the operations under their control” (Morse and Kimball 1951, p. 1). The U.S. military also adopted OR during the war, leveraging mathematical models and data to solve practical problems that were predictive and prescriptive in nature (Morse 1952).

After World War II, the concept of applying OR techniques to nonmilitary applications took root in both the United Kingdom and United States, leading to new university courses and degree programs. OR’s transition to nonmilitary applications was facilitated by industry cultures that were receptive to types of analysis used by the military OR teams, including the widespread use of scientific management and statistics used in industrial settings (Trefethen 1954). In the following decades, advances in theory, methodology, and application broadened the scope of OR and fostered the development of numerous university courses and degree programs.

OR has evolved since the 1950s, with modern definitions emphasizing its role in transforming data into actionable insights (INFORMS 2024a). At present, OR employs a variety of quantitative methods to analyze complex systems and make informed decisions. Table 1 contains a list of OR methods in use today. Classic methods include linear programming, integer programming, decision analysis, simulation, and queuing theory. Additional methods have been added over the decades to tackle a wide range of practical problems.

Table 1. OR Categories by Methods

With its advantages, OR has been used as a powerful tool that supports decision-making processes in many fields, including transportation and logistics, healthcare and social services, manufacturing and production, planning and optimization, and other areas. OR categories by application fields are summarized in Table 2. For a detailed discussion on each subfield within the categories mentioned above, readers are referred to Petropoulos et al. (2024). Note that ML has expanded the application domain of OR within these areas (to more prominently include prediction) and to other emerging areas such as retail and the sharing economy.

Table 2. Major Application Fields of OR

1.2. Contributions

OR has been a powerful tool supporting decision-making processes in various fields; however, its niche perception and limited accessibility hinder its broader adoption. This paper, motivated by the history of OR, aims to address these challenges by exploring how OR can be made widely accessible. To address this, we consider several related questions as follows. How do we address the mathematical skills gaps needed for OR professionals? How do we integrate OR with other advanced computational techniques, including ML, so the advantages of each are best utilized? How do OR professionals adapt to technological trends of AI/ML/big data to build more robust, scalable, and adaptable OR solutions? What should we do to cultivate an OR ecosystem of open-source tools, accessible resources, and active online communities? These are not new questions for the OR community. In a 1986 keynote talk, Herbert Simon advocated for greater collaboration between OR and AI, noting the many similarities between the fields (Simon 1987). More recently, three joint AI/OR workshops organized by INFORMS, the ACM Special Interest Group on Artificial Intelligence (SIGAI), and the Computing Community Consortium (CCC) in 2021, 2022, and 2023 have reignited this discussion and identified opportunities for scientific advances that would benefit from AI/OR collaboration (Das et al. 2022, Dickerson et al. 2023, Kulkarni et al. 2025). These workshops aimed to instigate impactful and innovative collaborations between the AI and OR communities. The final report recommends fostering collaborations between the communities through joint educational initiatives, interdisciplinary funding opportunities, and long-term research programs (Kulkarni et al. 2025). Furthermore, it emphasizes aligning academic incentives and publication standards to support cross-disciplinary work while also advocating for the development of real-world benchmarks and shared data sets.

In this paper, we build on these previous efforts and draw lessons from the global success of ML to introduce a set of recommendations to modernize OR’s outreach and integration strategies. As such, we present an action plan for increasing the accessibility and, ultimately, the popularity of OR consisting of 10 proposed actions across three areas: (1) community and public awareness, (2) educational outreach, and (3) research and technology. The actions involve various stakeholders inside and outside of OR.

This paper is structured as follows. Section 1 introduces the OR topic, its history, and application areas. Section 2 describes the different growth dynamics between OR and ML, pointing out opportunities for OR visibility and laying a foundation for the action plan in the following section. Section 3 proposes the action plan and discusses each recommendation involving stakeholders and strategies to better promote OR for accessibility. Finally, Section 4 concludes our paper and proposes a vision for expanding OR’s visibility and impact within the data science landscape.

2. ML and OR: Two Fields With Different Dynamics

ML is a rapidly evolving field within AI that focuses on developing algorithms capable of making predictions or decisions based on data. It encompasses, among others, predictive tasks, where the goal is to infer future outcomes, and generative tasks, where the focus is on creating new data resembling the input. ML has been defined as “the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous fashion, by feeding them data and information in the form of observations and real-world interactions” (Ng 2022).

Unlike traditional programming, where explicit instructions are provided, ML algorithms use statistical techniques to infer patterns and extrapolate trends from data. This data-driven approach makes ML suited to environments such as social systems, where an underlying model of system dynamics is absent or infeasible to construct. Additionally, ML is evolving beyond statistical predictions to training autonomous agents capable of reasoning and achieving their own goals, which presents both opportunities and risks. Whereas at present humans still excel at some complex tasks such as in long-term planning, rapid advancements in ML are reshaping the landscape of AI.

On the other hand, OR involves crafting mathematical models that represent real-world systems, often integrating constraints and objectives to determine optimal or near-optimal solutions. Originating from military strategy in World War II, OR has evolved to address complex challenges across diverse industries (see Table 2). It relies heavily on structured modeling, optimization techniques and expert domain knowledge, making it well suited to domains where the underlying mechanics can be explicitly modeled. Solution exploration techniques developed in the OR field are especially efficient at dealing with combinatorial spaces, thanks to the development of a variety of iterative heuristics and exact methods based on domain decomposition and pruning.

The differing trajectories of ML and OR reflect their inherent characteristics. ML’s growth has been propelled by advances in computing power, the availability of large data sets, and accessible open-source tools. Its applications, ranging from chatbots to recommendation systems, have brought it widespread visibility and adoption. OR, by comparison, has evolved steadily, driven by industry-specific needs in logistics, finance, and manufacturing. Although highly impactful, it is often perceived as a more specialized discipline requiring significant technical expertise. Whereas ML’s data-centric nature makes it adaptable to a broad range of fields, lowering entry barriers and enabling rapid adoption across sectors, OR is more reliant on domain expertise and tailored models, which can limit its accessibility. The following subsections delve further into the methodologies, applications, and potential synergies between ML and OR, forming the basis for some recommendations in Section 3.

2.1. Complexity and Visibility

Whereas OR has been instrumental in optimizing complex systems supporting decision-making processes, its applications are often specific to industries and might be highly complex, making it less visible to the general public. Many OR innovations operate behind the scenes, quietly optimizing systems such as airline scheduling, supply chain logistics, and traffic flow. In contrast, ML often offers “point solutions” to isolated components of systems that allow their impacts to be more immediately observed (Agrawal et al. 2022). Additionally, ML benefits from far greater visibility because of its direct impact on everyday life. From virtual assistants like Siri and Alexa to self-driving cars and personalized recommendations on platforms such as Amazon and Netflix, ML has become ubiquitous. Corporate marketing, media fascination, and public perception of ML’s “superintelligent” capabilities—such as generating art, exploiting hidden patterns, and predicting future events—further amplify its prominence. This creates a feedback loop: ML is widely adopted because it is highly visible, and its adoption amplifies its visibility.

Despite its successes, ML’s allure often obscures its limitations. ML usually provides fast and scalable solutions for automating individual tasks, but its power to redesign complex systems is regularly not realized (Agrawal et al. 2022). Real-world performance sometimes falls short of laboratory results, particularly when deployment data differ from training data or when rare events disrupt predictions. Overreliance on ML, fueled by its portrayal as a universal solution, has led to significant disillusionment in some cases, as seen in failures to generalize across settings or address ethical concerns (Christian 2021).

In comparison, OR’s applications, though critical for the world’s economies, receive less attention, despite an increasing number of OR publications utilizing ML (Subramanian and Teichgraeber 2023). Many of these contributions, such as wildfire management, real-time cancer therapy optimization, and disaster response, have a profound impact on society, even saving lives. OR has also played a vital role in optimizing biomanufacturing, vaccine logistics, and broader healthcare operations. Yet public recognition increasingly gravitates toward terms such as “artificial intelligence,” “generative AI,” or the emerging discourse on “agentic AI”—even in cases where impactful systems are fundamentally built on OR-based optimization. Beyond these high-stakes domains, OR also underpins many day-to-day activities, from optimizing bike-sharing systems to ensuring balanced resource allocation in emergency services. However, these contributions rarely drive paradigm-shifting narratives, making OR’s role in modern technology less visible. Unlike ML, which thrives on public fascination, OR’s work is often constrained to specialized applications, requiring expert knowledge and tailored problem-solving approaches.

Finally, the terminology associated with OR plays a role in shaping its visibility. The term “operations research” feels outdated and fails to capture the full breadth and scope of the field. Recent attempts to rebrand operations research as “predictive and prescriptive analytics” reflect efforts to modernize its image, but these changes have been slow to gain traction, and no clear consensus has emerged. In contrast, terms such as “machine learning” and “artificial intelligence” resonate more broadly and are inherently more appealing to the general public.

2.2. Hardware, Software, and Data Advances

OR traditionally relies on mathematical modeling and optimization techniques that can be computationally intensive, as does ML. However, tensor operations and neural network architectures used by ML algorithms map well to GPU-friendly computations. In contrast, classical OR methods such as mixed-integer linear programming (MILP) are challenging to parallelize because of their inherently sequential nature (e.g., branch and bound). Other emerging paradigms, such as quantum computing, show promise for both fields. Whereas their transformative potential is widely recognized, practical use remains farsighted.

ML benefits from rich, open-source ecosystems (e.g., TensorFlow, PyTorch, scikit-learn) and vast repositories of data sets (e.g., ImageNet, COCO). These resources allow researchers and practitioners to focus on introductory and applied tasks without a thorough understanding of the underlying algorithms. Open benchmarks for cutting-edge tasks, for example, in computer vision, structured reasoning, or sentiment analysis, further stimulate progress by providing standardized evaluation protocols and encouraging competition. By contrast, OR resources might be less accessible. Whereas classic OR problems have well-established data sets (e.g., MIPLIB, CVRPLIB, TSPLib), a large part of current cutting-edge research in OR increasingly focuses on dynamic and stochastic problems for which standard data sets and evaluation protocols are lacking. This absence limits reproducibility, hinders accessibility, and slows progress, though increased efforts are being realized in that direction (Tang and Khalil 2024). Moreover, some tools such as MILP solvers (e.g., Gurobi, CPLEX) are proprietary, raising some barriers for researchers and practitioners.

Foundational ML models, such as pretrained neural networks (e.g., BERT, GPT, ResNet), have become standard tools, lowering barriers to entry for nonexperts. OR lacks comparable frameworks. Declarative programming languages (e.g., AMPL, Pyomo) and solvers (e.g., Gurobi, CPLEX) are powerful but require significant domain expertise and have scalability constraints.

2.3. Interdisciplinary Nature and Funding

OR often requires collaboration across multiple disciplines (e.g., engineering, economics, management), which can hinder research and development efforts because of different methodologies and terminologies. By contrast, ML also benefits from an interdisciplinary foundation (drawing from fields such as computer science, neuroscience, psychology, and economics), but its ecosystem of tools and libraries often facilitates integration and broad adoption.

In terms of federal funding for basic scientific research in the United States, OR funding often falls within engineering, whereas ML research falls within computer and information science. Funding levels for engineering research in the United States are lower than for computer science relative to their sizes at many funding agencies (National Center for Science and Engineering Statistics 2024), which limits opportunities for OR breakthroughs and innovation, including those at the intersection of OR and ML.

Finally, OR’s traditional focus on industrial applications, although impactful to the world’s economies, may also limit its appeal compared with ML. ML’s visibility stems from transformative applications such as self-driving cars, virtual assistants, and generative AI, which capture public imagination. Consequently, ML has attracted substantial public and private funding, driven by its potential to revolutionize industries and create new markets. Examples such as OpenAI and Alphabet exemplify this trend. The ability of ML to scale horizontally across diverse applications, coupled with its perceived role as a driver of innovation, has made it a priority for venture capital and government funding. In contrast, OR funding is often tied to domain-specific challenges, though it still attracts support for high-impact applications, such as renewable energy systems, disaster response, and healthcare optimization.

2.4. Technology Transfer and Explainability

OR is seen as a mature field with established methods and applications, often perceived as part of the background infrastructure that keeps industries running smoothly. This mature status can lead to less media attention compared with emerging fields or a research funding imperative. By contrast, ML has captured public attention through its transformative potential and the hype and fear surrounding its societal impacts. Topics such as general AI, job automation, and the ethical implications of AI applications make ML a subject of both fascination and concern.

Explainability and transparency are critical challenges in both fields but manifest differently. In ML, the growing demand for trust and accountability has spurred the development of explainable AI (XAI) tools. These tools aim to interpret complex models, detect biases, and improve decision making. Despite significant advances, many explainability techniques in ML remain ad hoc, offering only “one possible” explanation, which can be easily manipulated, rather than a comprehensive understanding.

In comparison, explaining decisions and policies in OR remains a significant challenge, though the nature of this challenge differs from ML. Many OR algorithms, particularly those rooted in mathematical optimization, are inherently interpretable in theory. However, in practice, the scale, complexity, and constraints of real-world problems make it difficult to communicate the rationale behind optimization outcomes to stakeholders. For instance, the output of a mixed-integer programming model might be mathematically sound but difficult to justify intuitively, especially when trade-offs between conflicting objectives are involved. This perceived opacity can hinder adoption, though emerging approaches such as interpretable optimization and the development of stakeholder-friendly metrics start to address these issues.

2.5. Education and Skill Development

OR typically requires specialized knowledge in mathematical modeling, statistics, and optimization, which can be perceived as challenging and less accessible to a broader audience. Whereas formal OR courses are integrated into several academic programs, there are fewer widely available online resources compared with ML.

ML, on the other hand, has become a cornerstone of academic curricula and professional training programs. Data science degrees are increasingly popular in the United States, complemented by emerging certifications in fields such as data engineering, artificial intelligence, and ML-driven decision making (Pierson 2023). The abundance of online courses, boot camps, and other accessible resources has democratized ML education, enabling individuals from diverse backgrounds to join the field. Platforms such as Kaggle foster active learning through code sharing, data sets, and competitions, creating a global community of practitioners and researchers. Government initiatives, such as centralized data repositories promoted by the White House, further support AI education and training across various domains (The White House 2023).

Efforts to bridge the skill gap and foster interdisciplinary collaboration between OR and ML are gaining traction. For instance, a 2024 summer school program trained PhD students in OR and computer science to integrate methodologies from both fields. This initiative aimed to prepare the next generation of researchers to develop foundational tools and tackle cross-disciplinary challenges.¹ Similarly, the EURO-NeurIPS challenge (Kool et al. 2022) united the OR and ML communities to solve a dynamic vehicle routing problem with time windows, using real-world data provided by ORTEC. At the same time, many university OR departments have introduced courses on ML to train students in topics at the intersection of ML and decision making (e.g., see Boutilier and Chan 2023). Interdisciplinary workshops and university courses help bridge terminology differences between the fields and can enable students to develop complementary skills make advances in both fields.

2.6. Publication Practices

The OR community maintains a “fascination toward optimality,” which serves as both an asset and a challenge. On one hand, well-defined metrics enable rigorous evaluation of solution methods on standardized problem instances. On the other hand, an overemphasis on achieving optimality can limit the practical relevance of research, slowing the industrial transfer of results. In many real-world settings, approximate or heuristic solutions often suffice and are more implementable. Furthermore, gatekeeping in top OR venues often encourages researchers to concentrate on stylized problems where strong theoretical results are attainable. Whereas this focus strengthens the field’s mathematical foundations, it also creates a disconnect from the stochastic and complex nature of real-world situations.

In contrast, the ML community prioritizes empirical research, application success, and faster publication turnover through conferences. Leading venues such as the Conference on Neural Information Processing Systems (NeurIPS), the International Conference on Machine Learning (ICML), and the International Conference on Learning Representations (ICLR) emphasize state-of-the-art performance on benchmarks and real-world tasks, accelerating the transfer of ML innovations to industry and fostering rapid progress. However, this empirical focus has also raised concerns about reproducibility, lack of theoretical rigor, and dependence on computationally intensive methods. Whereas methodological breakthroughs remain highly prestigious, the recognition of empirical contributions keeps ML research closely aligned with practical needs. The length of review times also differs in the two communities, with long review times for many prestigious OR journals that may dissuade those working at the intersection of OR and ML from pursuing publication in OR journals. Additionally, promotion and tenure criteria in many OR departments tend to undervalue conference publications, creating institutional disincentives that hinder the cross-fertilization of ideas between the two fields.

The evaluation practices in OR and ML also differ significantly. In OR, researchers typically adopt a decomposition mindset, separating the task from the mathematical model and method. This approach offers a clearer understanding of two distinct sources of imprecision: (i) the performance of algorithms solving the model, often assessed through metrics such as optimality bounds; and (ii) the adequacy of the model itself for representing the given task (see, e.g., the discussion in Gribel and Vidal 2019). By contrast, ML performance evaluations are typically task oriented, directly measuring outcomes. Whereas this approach closely ties methodological progress to real-world performance, it aggregates multiple factors into a single evaluation metric, potentially obscuring the sources of error or limitations.

3. How to Make OR More Accessible?

Promoting the popularity of OR requires well-planned strategic actions and coordinated efforts among multiple stakeholders. Six relevant stakeholders are identified and represented with their logos along 10 proposed key action plans:

Academic institutions of higher education, including professors and administrators;

Elementary and secondary school (K-12) academic institutions, including teachers, guidance counselors, and leaders;

Professional associations, such as INFORMS, IFORS, and AAAI;

Industrial organizations and corporate partners;

Government policymakers, legislators, and state accreditation bodies;

Media, including media organizations and science journalists.

In the following subsections, we present 10 action plans designed to promote OR to a broader audience. The 10 strategies, developed based on the analysis and insights from Section 2, are organized into three broad categories: community and public awareness, educational outreach, and research and technology.

3.1. Community and Public Awareness

Building a robust community network is crucial for achieving success and ensuring sustainable development. Strengthening professional networks and organizations is key to offering resources (Action 1), networking opportunities, and support for OR professionals. By sharing best practices (Action 2), providing strategic direction (Action 3), and assisting with administrative efforts, local professional chapters can become more dynamic and engaged. Additionally, organizing and promoting OR-focused events fosters a vibrant community and attracts new talent to the field.

Table

Table

Table

To successfully implement the suggested actions, several tasks are involved, such as public education campaigns and media presence. Through public education campaigns, OR can be promoted by highlighting its importance and applications in everyday life. Various case studies, success stories, and real-world examples can be used to demonstrate its impact in consequential applications. INFORMS recognizes outstanding examples of OR and advanced analytics with the Franz Edelman Award. Many of the finalist projects in recent years have used a combination of multiple analytical approaches, including OR and AI/ML. Highlighting and showcasing the Edelman Award finalists in more public venues could be the cornerstone of public education campaigns. Furthermore, increasing the exposure in mainstream media by publishing articles, participating in interviews, and featuring success stories in popular science and business magazines can expand the visibility of OR. Recent efforts, such as the INFORMS Advocacy Initiative and the media training made available at the 2022 INFORMS Annual Meeting, represent positive progress toward this goal.

Moreover, as discussed in the previous section, the OR term is somewhat outdated, so it is important to consider rebranding it with a modern and dynamic image that resonates with younger generations and the tech-savvy and broader AI community. Clear and compelling messaging should be developed to explain what OR is, its benefits, and how it differs from and complements other fields such as ML.

3.2. Educational Outreach

Education enriches society by broadening opportunities and fostering a more informed, skilled population. Achieving this requires integrating OR topics into standard curricula at the K-12 (Action 4), and undergraduate levels (Action 5), which will help expose a broad audience to its concepts and applications. Developing engaging courses and workshops can further introduce students to optimization early in their education.

Accessible online resources, such as courses, tutorials, and webinars on platforms like Coursera, edX, and Khan Academy, can extend OR’s reach to a global audience. Additionally, hackathons and competitions focused on practical applications and decision making offer dynamic ways to engage students and professionals. Promoting OR career pathways within educational and workforce development programs can attract talent and build interest while educating the public on how optimization influences everyday life and decision-making processes.

To sustain OR’s relevance and impact, it is essential to invest in strengthening its talent pipeline. OR must be integrated into K-12 curricula—not as an afterthought but as a core component of math and data science literacy. At the same time, bold initiatives such as “OR for All” microcertifications—short, accessible, and hands-on training programs—are urgently needed to democratize OR skills for students and professionals alike. Without a deliberate, accelerated strategy to cultivate OR-savvy talent, the field risks falling behind in critical areas such as supply chains, public health, and sustainability.

Table

Table

An increasing number of innovations and scientific breakthroughs have recently been driven by industry leaders such as Facebook/Meta, Apple, Amazon, Netflix, and Google/Alphabet (FAANG). It is essential to collaborate with industries (e.g., starting with flagship firms) to demonstrate successful OR applications and their impact on business efficiency, cost savings, and innovation. Publishing more case studies and white papers could spur more collaboration. Additionally, offering professional development programs and certifications in OR for industry professionals is crucial. Partnering with industry leaders to create internships and co-op programs can also bridge the gap. Moreover, establishing a consortium platform where academia, public, and private sectors collaborate can harness the strengths of each stakeholder to achieve common goals (Action 6).

Table

3.3. Research and Technology

Leveraging technology has the potential to emphasize how OR utilize big data and advanced analytics to solve complex problems as well as to position it as a complementary field to ML in data-driven decision making (Action 7). Developing and promoting user-friendly OR software tools can empower practitioners to apply optimization techniques without requiring deep technical expertise, thereby democratizing the use of decision-making tools and broadening its user base. Key to success is in lowering the barrier of entry for users without advanced degrees in OR to engage with AI. Examples that have achieved this include the INFORMS Insights webinar series (INFORMS 2024b) and courses designed to make AI accessible to undergraduates in OR (see Albert 2025).

Table

Table

Efforts should focus on fostering collaboration between OR and ML/AI communities by demonstrating how combining them can solve more complex and multifaceted problems (Action 8). Cross-disciplinary research projects and conferences should be encouraged to bring together experts from OR, ML, data science, and other related fields (Action 9).

ML should be viewed as an “enabler” and not a competitor. For example, MILP formulations require a good mathematical background as well as extensive practice to make good choices. ML tools might help us in the near future to automatically convert a problem defined in a natural language into a mathematical formulation (Ramamonjison et al. 2023). ML is also already effective at algorithmic selection and parameter fine-tuning, a task that often represents a challenge in OR applications.

There have long been synergies between AI and OR professional communities. In particular, AI has been advanced by foundational computational tools relying on optimization from the OR community. However, research communities have just started to tap into the potential of new optimization-based approaches designed for the application of AI. Integration between AI and optimization offers the potential to advance algorithmic decision making.

Potential synergies between AI and optimization could help solve problems at the scale and complexity demanded by real-world applications of AI. Areas that are fertile for exploration could (1) combine probabilistic planning and reasoning in AI with modeling, stochastic programming, and probability models from the OR community, (2) algorithmic techniques from the optimization community to tackle interpretability, trustworthiness, and fairness often desired in AI decision-making settings, and (3) techniques that exploit massive data sets (from AI) with abilities (from OR) to extract features using model-based approaches to drive decision making (Dickerson et al. 2023, U.S. National Institute of Standards and Technology 2023).

In turn, many core challenges in trustworthy ML (e.g., connected to audit and robustness checks) require formal guarantees of performance, and ad hoc heuristic approaches often fail to deliver consistent and reliable solutions. For instance, counterfactual explanations for nonconvex ML models often fail to identify the most natural or informative explanations (Parmentier and Vidal 2021), so exact optimization models would be desirable. Similarly, robustness against edge cases and privacy properties (Ferry et al. 2024) must be formally verified in critical systems to ensure safety and reliability, whereas regulations might soon impose more stringent guarantees regarding model equity, which might require mathematical proving capabilities. For all these situations, OR-based methods can make significant contributions.

Realizing this vision also requires institutional adaptation. University policies must evolve to recognize contributions at the AI-OR interface. It is crucial that university promotion and tenure requirements in OR departments adopt policies that give publications in AI venues full consideration for their contributions. Many OR departments, including those of the authors, have adopted this practice for faculty members performing research in these areas.

Table

Finally, sustained investment is vital to support research and development in OR. Increased funding for OR research from government agencies (NSF, DoE, DoT, NIH, DoD/DARPA, etc.), private foundations, and industry grants should be advocated by emphasizing the economic and societal benefits that breakthroughs in OR can have (Action 10). Highlighting the commercial potential of optimization solutions and benefits could encourage venture capital investment in OR-focused startups and entrepreneurial ventures. Likewise, writing and distributing white papers to funding agencies that identify and outline specific areas of research, especially at the intersection of OR and ML, may be an effective avenue for advocating for OR research. The three joint AI/OR workshops organized by INFORMS, ACM SIGAI, and the CCC produced reports that emphasize this point and identify opportunities for scientific advances at the intersection of AI and OR (Das et al. 2022, Dickerson et al. 2023, Kulkarni et al. 2025).

Table

4. Conclusions

This paper emphasized the pressing need to make OR more accessible and widely utilized, particularly given its extensive history and established methodologies across optimization, decision analysis, and systems engineering. Unlike ML, OR has struggled with visibility, limited technological adoption, and public awareness, which hampers its accessibility and impact. Through a comparison with ML, we identified barriers such as limited open-source (e.g., data, codes, and tools) availability, less interdisciplinary collaboration, and lower funding opportunities that have restricted its broader adoption and integration.

Addressing these barriers requires multifaceted strategies. Increasing OR’s visibility and public awareness can shift perceptions, highlighting its relevance in contemporary problem solving alongside ML. Expanding educational outreach and creating accessible training resources can demystify optimization methodologies, preparing a new generation of researchers and practitioners with mathematical knowledge and practical skills. Leveraging technology advancements, including data accessibility and intuitive software, can streamline OR processes, making its tools easier to adopt across sectors. Further, fostering synergies between ML and OR by integrating ML’s predictive capabilities with OR’s optimization strengths could drive powerful, hybrid applications. Strategic partnerships with industry, bolstered by funding and investments, can promote practical applications of OR, demonstrating its value in real-world decision making and problem solving. Building concrete action plans along with identifying corresponding key stakeholders can help the field gain the kind of mainstream recognition ML has achieved by clearly connecting its capabilities with real-world applications and societal benefits.

Ultimately, making OR more accessible will not only enhance the field itself but also open up new avenues for interdisciplinary research, expand its application across diverse industries, and drive impactful innovations that address complex global challenges. With strong stakeholder commitment and sustained effort, we can achieve key milestones in medium term—including demystifying OR through real-world case studies (action 2), launching academic awareness campaigns (action 3), fostering flagship industry collaborations (action 6), promoting OR-ML interdisciplinary research (action 9), and securing dedicated funding (action 10). Meanwhile, foundational initiatives such as establishing an open-access OR knowledge hub (action 1), reforming STEM education pipelines (action 4), and integrating mandatory OR-AI training (action 7) require longer-term investment but are critical for enduring transformation. In about a decade, OR could stand as the backbone of decision making in every sector—if key stakeholders, funding agencies (NSF, NIH, DARPA), and academic (industrial engineering (IE)/industrial and systems engineering (ISE)) departments prioritize accessibility, interdisciplinary collaboration, and public engagement. The time for incremental change has passed; the future of OR demands bold, visible action today.

Overall, OR methodologies and software provide unique value, and the proposed 10 actions will enhance their accessibility to the public. However, several critical questions must be addressed to fully realize OR’s potential.

1. Openness and accessibility must become a priority within OR. OR has traditionally emphasized algorithmic sophistication, formal guarantees, and rigorous mathematical foundations—capabilities that remain crucial, especially in domains where transparency, reliability, and accountability are nonnegotiable. These strengths are often complementary to those of ML, which tends to prioritize scalability and data-driven performance. However, whereas OR’s theoretical strength remains an asset, it is not sufficient on its own to ensure broad adoption. A major reason behind ML’s rapid growth lies in the accessibility of its ecosystem: widely available open-source libraries (e.g., TensorFlow, PyTorch, scikit-learn) and vast public data sets have lowered entry barriers and enabled rapid prototyping and learning. In contrast, OR tools and resources often remain fragmented or proprietary, hampering experimentation, learning, and real-world adoption. To thrive in today’s research and application landscape, OR must pair its formal rigor with greater accessibility through community standards, open benchmarks, and especially requirements to share both code and data upon publication. Ultimately, ease of access can be more influential in a field’s growth than methodological sophistication alone. For OR to retain its relevance and expand its reach, it must champion both.

2. OR’s long-term relevance depends on earlier, broader education. OR education has traditionally focused on graduate-level training, with most exposure coming through specialized master’s programs or doctoral-level research. Even joint AI-OR workshops and summer schools typically cater to PhD students and early-career researchers. But that’s too late. By that stage, many students have already chosen their academic and professional trajectories. To broaden OR’s influence, we must remove barriers and create opportunities for early engagement, particularly at the K-12 and undergraduate levels. Introducing foundational OR concepts—such as optimization, decision making under uncertainty, and trade-off analysis—early on can spark curiosity and make the field more visible and approachable. The counterintuitive insight is this: ensuring OR’s long-term impact depends less on cultivating a small number of elite experts and more on demystifying its ideas for a broad, young, and diverse audience.

3. The OR publication model must evolve to meet modern research needs. The OR community has traditionally centered its publishing around top-tier journals, which, although rigorous, often entail multiyear review and resubmission cycles. Reviewers commonly request major redesigns of papers beyond sole technical verification—often shaped by their own stylistic and methodological preferences—which can significantly slow down dissemination and innovation. By contrast, ML’s leading venues operate on strict, fast-paced conference cycles. Reviews are conducted within weeks, based on the paper “as is,” and authors receive final decisions with minimal iterations. This system is not without drawbacks: it limits dialogue and revision, and review quality can be less consistent. Still, it promotes timely evaluation. To remain competitive and relevant, OR should progress toward shorter review cycles and more agile dissemination models, particularly to attract interdisciplinary work at the OR-AI interface. INFORMS journals have made notable efforts in this direction. However, a broader paradigm shift is required to fundamentally reshape review culture—preserving rigor without exhausting authors’ time. Alternatively, OR could pursue a more structural transformation by fostering the growth of large, selective conferences with full proceedings, akin to NeurIPS, ICML, or ICLR in ML. For such a model to succeed, institutional norms must evolve—universities must value these venues appropriately, and authors must embrace them as legitimate first-tier options. ML has a clear rhythm: three flagship conferences, three submission cycles, and three predictable review timelines per year. OR, by contrast, still lacks a shared answer to a deceptively simple question: what are the top three conferences in our field?