Prescriptive Analytics

Analytics comes in three levels: descriptive analytics reports what happened, predictive analytics forecasts what will happen, and prescriptive analytics recommends what to do. INFORMS, the professional society for operations research and analytics, lays out exactly this three-part taxonomy, with prescriptive analytics helping decision making by providing actionable advice.

The machinery differs by level, as description runs on reports and dashboards, prediction runs on statistical and machine learning models, and prescription runs on optimisation, simulation, and decision rules that weigh constraints and trade-offs. The output is a decision, not a chart.

A prescriptive system for delivery routing does not predict traffic; it tells each driver which route to take, given the traffic prediction. Real deployments stack the levels, so a retailer predicts demand per store, then prescribes inventory allocations that minimise stockouts against holding costs, using the forecast as an input to the optimiser.

The limits are the models' limits, since an optimiser faithfully maximises whatever objective it is given, so a badly chosen objective, like cost without service penalties, produces confidently wrong recommendations at scale. Data quality gates everything downstream, because an optimiser fed garbage forecasts produces confident, precise, wrong recommendations whose polish makes the errors harder to spot than a bad spreadsheet ever was.

Adoption fails most often on trust: managers overrule black-box recommendations, so successful systems expose their reasoning and let humans adjust constraints rather than replacing judgement wholesale. Change management consumes most project budgets for a reason, since planners who spent years honing intuition need a system that negotiates with them, showing trade-offs and accepting overrides, rather than one that issues orders.

The discipline is older than the buzzword: operations research teams have prescribed decisions since the Second World War, when convoy routing and bombing accuracy studies founded the field, and modern computing simply put the same mathematics on every manager's desk. The economics justify the effort when decisions are frequent and similar.

Thousands of daily choices about prices, routes, or stock levels compound small improvements into serious money, which is why airlines, retailers, and logistics firms adopted prescription first. One-off strategic choices are a poorer fit, because optimisation thrives on repetition and feedback, and a merger decision happens once and the data it needs barely exists.

For a non-finance manager, the framing question for any analytics proposal is which level it serves. If the vendor shows you predictions, ask what decision they drive, because a forecast nobody acts on is an expensive horoscope.