Prescriptive analytics

Prescriptive analytics goes one step past prediction: instead of just estimating what is likely to happen, it recommends what to do about it. Given a forecast, a set of constraints, and an objective—maximize margin, minimize stockouts, hit a service level—it outputs a specific action or set of actions rather than a number.

Techniques include optimization (linear and mixed-integer programming), simulation, decision trees, and rule engines, often layered on top of predictive analytics or machine learning outputs. Some prescriptive systems also draw on causal inference to estimate what would happen under an action the organization hasn't yet taken, rather than just what has correlated with outcomes in the past.

It matters because a forecast alone doesn't tell a planner how many units to reorder or which customers to prioritize for retention outreach—prescriptive analytics closes that gap between insight and action. It is increasingly embedded in augmented analytics tools that surface a recommended next step alongside a chart. The main pitfall is opacity: a recommendation is only trustworthy if its constraints and objective function are visible and can be challenged, since an optimizer will confidently produce a well-optimized answer to the wrong problem if it is given a flawed objective.