Operations research analysts

Three parts of your work where AI is already doing real lifting, and what stays yours.

• Empirical field observation + probabili…1937–1947

  Empirical field observation + probability tables (pre-computing OR)

  The first operational research work used no computers: Blackett's teams at Coastal Command collected combat reports by hand, built frequency tables of U-boat sighting distances and patrol aircraft tracks, and applied classical probability and statistics from paper tables. The tools were slide rules, actuarial tables, and systematic observation forms. The insight was methodological: that disciplined data collection and mathematical analysis of operations in progress could improve outcomes faster than intuition or tradition alone. This era established the fundamental OR discipline of treating operational problems as quantitative decision problems, regardless of whether a computer was available to solve them.

• Linear programming + mainframe batch co…1947–1965

  Linear programming + mainframe batch computing (Dantzig simplex, IBM 701/7090)

  George Dantzig's simplex algorithm, developed at the Pentagon in 1947, was the first general-purpose computational method for finding optimal solutions to constrained resource allocation problems. It made linear programming tractable for real industrial problems: the Air Force's Project SCOOP (Scientific Computation of Optimal Programs) tested the simplex on IBM mainframes from 1952. By the late 1950s, RAND Corporation and major defense contractors were solving LP problems with hundreds of variables on IBM 701 and 7090 machines. The key constraint was batch computing: an OR analyst would submit a deck of punched cards overnight and receive results the next morning. Model development was slow, debugging was laborious, and computational scale was the binding limit. Airlines (American Airlines' SABRE reservation and yield-management precursors), oil refineries (refinery scheduling was among the earliest commercial LP applications), and defense logistics all benefited from this era.

  Effect on the work

  Linear programming on mainframes enabled problems to be solved that previously required armies of clerks doing iterative manual calculations. A single OR analyst with LP model access could solve scheduling and allocation problems that had previously been either intractable or left to managerial rule-of-thumb. The era grew OR as a profession rather than displacing any existing role.

• Interactive computing + commercial LP s…1965–1985

  Interactive computing + commercial LP solvers (IBM System/360, LINDO, MPS/360)

  IBM's System/360, introduced in 1964, and the shift toward interactive time-sharing computing in the late 1960s and 1970s fundamentally changed how OR analysts worked. LINDO (Linear Interactive and Discrete Optimizer), developed at the University of Chicago in 1980, democratized LP by running on minicomputers. The OR analyst's role shifted from supervising batch runs to interactively building and testing models. Simulation emerged as a parallel tool set (GPSS, SIMSCRIPT), and network flow algorithms (Ford-Fulkerson, Dijkstra) enabled routing and scheduling problems to be solved at new scale. This era also saw OR expand beyond defense into healthcare scheduling, telecommunications network planning, and early financial optimization.

• Personal computing + spreadsheet modeli…1985–2000

  Personal computing + spreadsheet modeling (Lotus 1-2-3 1983, Excel 1985, What's Best! / @RISK)

  The IBM PC and the spreadsheet transformed OR from a specialist mainframe activity into something a manager with a laptop could attempt independently. Lotus 1-2-3 (1983) and Excel (1985) put linear programming add-ins (What's Best! for LP, @RISK for Monte Carlo simulation) into the hands of decision-makers who had never written Fortran. This democratization was a double-edged development for OR professionals: it expanded the market for analytical thinking throughout organizations while simultaneously reducing the barrier to entry for basic optimization and simulation work. The professional OR analyst's comparative advantage shifted toward complex multi-period, multi-constraint, stochastic models that Excel could not handle, and toward the interpretation and deployment work that spreadsheet models left untouched.

• Commercial solver platforms + supply ch…1997–2015

  Commercial solver platforms + supply chain digital twins (Gurobi 2008, CPLEX, SAP APO, Kinaxis)

  The late 1990s and 2000s brought solver technology to its modern form. CPLEX (acquired by IBM in 2009) and Gurobi (founded 2008 by former CPLEX developers) became the benchmark MIP solvers, capable of solving problems with millions of variables in minutes that would have taken days on 1980s hardware. Simultaneously, enterprise supply chain planning platforms (SAP Advanced Planner and Optimizer, Kinaxis RapidResponse, JDA) embedded OR optimization into business processes: an OR analyst at a major manufacturer no longer needed to write solver code from scratch but instead configured and validated platform models and interpreted their outputs for operations teams. Python (SciPy, PuLP) arrived as the coding platform that replaced Fortran and COBOL in custom OR work. The discipline professionalized: the INFORMS Certified Analytics Professional (CAP) credential launched in 2013 to distinguish professional OR practitioners from generalist analysts.

  Effect on the work

  The platform era significantly expanded who could use optimization: supply chain managers without OR training could operate solver platforms as configured tools. This did not reduce OR analyst employment; it expanded the scope of optimization adoption fast enough to grow the workforce substantially. The BLS OEWS series for 15-2031 shows steady employment growth from 2000 onward.

• Python ML stack + cloud compute2015–2022

  Python ML stack + cloud compute (scikit-learn, PyTorch, AWS/Azure, Jupyter notebooks)

  Python's ecosystem for scientific computing matured in the 2013-2016 period (NumPy, SciPy, pandas, scikit-learn, and then PyTorch) and cloud compute made large-scale optimization and simulation economically accessible to organizations that could not afford dedicated HPC clusters. The OR analyst role became a hybrid: traditional prescriptive optimization (LP, MIP, simulation) combined with machine learning inputs (demand forecasting, anomaly detection, price elasticity modeling) that fed optimizer parameters. The job title landscape fragmented: "Data Scientist," "Decision Scientist," "Applied Scientist," "Analytics Engineer," and "Planning Analyst" all described roles that, in an earlier era, would have simply been called operations research. Many practitioners working in logistics, pricing, revenue management, and supply chain did not formally identify as OR analysts even though their work was OR in substance.

• LLM-augmented optimization2022–present

  LLM-augmented optimization (Cursor + Claude Code for PuLP/Gurobi scaffolding, AI supply chain platforms)

  GPT-4 (March 2023) and purpose-built coding assistants like Cursor (2023) and Claude Code (2025) changed how OR analysts build models: a natural-language description of a vehicle routing problem or a workforce scheduling challenge can generate a first-pass PuLP, OR-Tools, or Gurobi formulation in minutes, compared to hours of manual coding. Kinaxis Maestro (launched 2024) and o9 Solutions embedded generative AI into supply chain planning platforms, producing AI-recommended re-plan actions in response to disruption scenarios and generating natural-language explanations of recommended changes. Gurobi documents that LLMs and mathematical solvers are complementary: LLMs contribute fluency and problem articulation while solvers provide the provable precision optimization requires, with human-AI-solver loops emerging as the new standard workflow. The net effect is that junior-tier formulation work is now substantially AI-augmented, while the irreducibly human tasks (problem framing with operating stakeholders, model validation against physical reality, deployment change management) are becoming more rather than less valuable as AI-generated models proliferate and require expert review.

  Effect on the work

  BLS projects employment growth for 15-2031 in the 2024-2034 cycle, the fastest-growing outlook the occupation has ever received. The profession is growing into AI augmentation rather than shrinking from it: organizations that previously could not afford full-time OR staff can now access OR-quality optimization via platforms, while organizations that already had OR capacity are expanding the scope of problems they attack.