Why Operations Research Still Matters in the Age of AI

The problem with "just use AI"

Every week, a new tool promises to optimize your operations with AI. Feed it your data, let the model figure it out, and watch efficiency improve. The pitch is compelling. The reality is more complicated.

Most operational problems are not prediction problems. They are decision problems — and the distinction matters. Predicting demand is useful. Deciding how to allocate a mixed fleet across 400 deliveries with time windows, capacity limits, and cost constraints is a fundamentally different challenge. The second problem has structure, and that structure responds to methodology.

What operations research actually does

Operations research (OR) is the discipline of applying mathematical and analytical methods to decision-making. It includes linear programming, integer optimization, constraint satisfaction, network flow models, scheduling theory, and simulation.

These are not new techniques. They have been used for decades in logistics, manufacturing, energy, telecommunications, and defense. What is new is how accessible they have become — and how few companies outside of large enterprises take advantage of them.

The core value of OR is this: it finds the best decision given a set of constraints. Not an approximate answer. Not a pattern learned from historical data. The mathematically optimal — or near-optimal — decision for the specific situation at hand.

Where OR outperforms general-purpose AI

Consider a delivery routing problem. A machine learning model trained on historical routes might produce reasonable plans for typical days. But when constraints change — a vehicle breaks down, a new delivery window is added, a depot closes — the model has no mechanism to reason about feasibility. It can only interpolate from what it has seen.

An optimization model, by contrast, encodes the constraints explicitly. It knows what is feasible and what is not. It can produce a new plan in minutes that respects every operational constraint while minimizing cost or time. When the problem changes, the model adapts — because it reasons from structure, not from patterns.

This does not mean AI has no role. Prediction models are valuable for demand forecasting, anomaly detection, and classification tasks. But when the core challenge is making a decision under constraints, operations research is the right tool.

Why it remains underused

Three reasons stand out:

1. It requires domain modeling. OR is not a plug-and-play tool. You need to understand the problem well enough to formulate it mathematically — which means understanding the business, the constraints, and the objectives. This takes time and expertise.

2. It does not market well. "We built a mixed-integer linear program" does not generate the same excitement as "We built an AI." The substance is there, but the narrative is harder to sell.

3. The talent gap is real. Few software teams have experience with optimization modeling. Most developers can build a CRUD application or integrate an API, but formulating a vehicle routing problem as a MILP model requires a different kind of thinking.

The opportunity

For companies with operational complexity — logistics, manufacturing, field services, energy, healthcare — operations research represents a significant and often untapped source of efficiency. The improvements are not marginal. In our own work, we have seen 15-25% cost reductions from optimization models applied to delivery planning.

The companies that invest in this discipline will have a structural advantage: better decisions, lower costs, and systems that adapt to changing conditions — not because they adopted the latest trend, but because they applied the right methodology to the right problem.