Prescriptive Analytics is the use of analytical models, optimisation methods, simulations, rules and related techniques to determine or recommend actions that could produce desired outcomes under specified objectives and constraints. Whereas Descriptive Analytics explains what has happened or what is currently observed and Predictive Analytics estimates what may happen, Prescriptive Analytics addresses questions about what action should be taken within a defined decision context. It can be used for resource allocation, scheduling, pricing, logistics, inventory decisions, workforce deployment, capacity planning and other problems where there are multiple possible actions and measurable consequences. The word prescriptive reflects its focus on action rather than observation alone.
Prescriptive Analytics normally requires three interconnected elements: an understanding or model of the relevant system, a set of possible decisions and an objective or set of objectives against which alternatives can be evaluated. The model may incorporate forecasts, probabilities, costs, capacities, constraints and business rules. An optimisation algorithm can then search for a combination of decisions that maximises or minimises a defined objective. For example, a logistics problem might attempt to minimise transport cost while satisfying delivery requirements and vehicle capacities. The resulting recommendation is meaningful only because the objective and constraints have been explicitly defined.
Optimisation is a major component of Prescriptive Analytics. Linear programming, integer programming, nonlinear optimisation, constraint programming, dynamic programming and heuristic methods can be used depending on the structure of the problem. Exact optimisation methods seek mathematically optimal solutions under a specified model, while heuristics and metaheuristics can find good solutions when exact computation is impractical. Simulation can evaluate possible policies under uncertainty, and scenario analysis can compare alternative decisions. The choice of method depends on problem size, mathematical structure, uncertainty, computational constraints and the required decision speed.
Prescriptive Analytics often relies on Predictive Analytics because recommendations may require estimates of future demand, risk or other uncertain outcomes. A retailer might forecast demand and then optimise inventory or replenishment decisions based on those forecasts. However, prediction alone does not determine the optimal action because the organisation may face constraints involving budget, capacity, service levels, labour availability or risk tolerance. Prescriptive systems therefore combine expected outcomes with decision variables and constraints. Errors in predictions can propagate into recommendations, making sensitivity analysis and robust decision-making important.
Objectives can be single or multiple. A decision model might minimise cost, maximise profit, reduce delivery time, improve service reliability or balance several competing goals. Multi-objective optimisation can expose trade-offs rather than producing one universally optimal answer. Organisations may use weights, constraints or priority structures to represent preferences among objectives. These choices are not merely technical because changing the objective function can change the recommended decision substantially. Prescriptive Analytics therefore requires meaningful involvement from decision-makers who understand organisational priorities and the consequences of different trade-offs.
Uncertainty presents a significant challenge. Traditional optimisation may assume known values for demand, costs or capacities, but real-world conditions often vary. Stochastic optimisation, robust optimisation and simulation-based approaches can incorporate uncertainty in different ways. A solution that is mathematically optimal under one forecast may perform poorly if demand changes substantially. Robust approaches may deliberately choose solutions that perform acceptably across a range of scenarios rather than maximising performance under one expected scenario. The appropriate approach depends on the cost of uncertainty and the organisation's tolerance for risk.
Implementation requires more than producing a mathematical recommendation. The organisation must have the operational ability to execute the decision, and the recommendation must be understandable and compatible with policies, systems and human responsibilities. A scheduling model that produces an optimal timetable may be unusable if employees cannot work the required shifts or if critical operational information is missing. Human decision-makers may also need authority to override recommendations when exceptional circumstances arise. Monitoring actual outcomes against model assumptions can reveal when a decision model requires revision.
Prescriptive Analytics differs from predictive modelling because it explicitly addresses actions rather than merely outcomes. It also differs from ordinary business rules because optimisation can evaluate combinations of decisions and trade-offs across a larger decision space, although rules and optimisation can be combined. Its benefits include systematic decision support, improved resource allocation and explicit treatment of constraints, while its limitations include model complexity, dependence on assumptions, data requirements and difficulty representing all relevant human or organisational factors. A prescriptive recommendation should therefore be understood as a model-dependent decision aid whose quality depends on the objective, assumptions, data, constraints and implementation environment.
