From Forecasts to Decisions: What Prescriptive Analytics Actually Does

Descriptive analytics explains what happened. Predictive analytics estimates what is likely to happen next. Prescriptive analytics goes a step further and answers a different question: given what's likely to happen, what should be done about it? That distinction sounds subtle, but it changes what the output of an analytics system looks like — not a chart or a forecast, but a recommended action, ranked against the alternatives, with the trade-offs made explicit. It's also the step where analytics stops being purely analytical and starts running into the operational constraints of the organization using it.

§ 02The components of a prescriptive system

A prescriptive system is less a single algorithm than a pipeline of several distinct components working together. Forecasting models — often the same predictive models used in descriptive analytics — estimate the range of outcomes that are plausible given current conditions. Optimization methods, drawn from a toolbox that includes linear and mixed-integer programming as well as heuristic and evolutionary algorithms, search across the space of possible decisions for the ones that perform best against the forecast, subject to whatever constraints the real world imposes — budgets, capacity, regulations, time. Simulation plays a supporting role, stress-testing a recommended decision against scenarios the optimization didn't explicitly consider, before that recommendation reaches a person who has to act on it.

§ 03Where the same pattern recurs across industries

The specific decisions differ by industry, but the underlying pattern repeats: a resource is scarce, demand for it is uncertain, and the cost of getting the allocation wrong is asymmetric. A hospital allocating operating rooms, a retailer setting inventory levels across stores, a utility balancing supply and demand on a power grid, and a logistics provider planning vehicle routes are all instances of the same structure — forecast the uncertain side, optimize the allocation of the scarce resource, and simulate before committing. What changes between industries is the shape of the constraints: regulatory limits in healthcare and finance, physical capacity in manufacturing and energy, time windows in logistics. Recognizing the shared structure is often more useful than treating each industry's version as a bespoke problem.

§ 04What makes these systems hard to build and maintain

The conceptual framework is straightforward; the implementation is not. Forecasts are only as good as the data feeding them, and in many organizations that data is incomplete, inconsistent across systems, or simply not collected at the granularity the optimization needs. Building and validating the optimization models themselves requires a combination of domain expertise and technical skill that's in short supply, and the systems are not cheap to build or maintain. Perhaps the largest barrier isn't technical at all: a prescriptive system that recommends a different way of working only creates value if the people who do that work adopt the recommendation, and getting from "the model suggests X" to "the organization does X" is an organizational change problem as much as an analytical one.