Data Analytics: Prescriptive Analytics and Optimization

While descriptive analytics tells you what happened and predictive analytics forecasts what might happen, prescriptive analytics determines what you should do. This advanced field uses optimization and simulation models to recommend the best course of action among various alternatives, given complex constraints and business objectives. For business leaders, mastering prescriptive analytics transforms data from a reporting tool into a direct advisor for strategic decision-making, optimizing everything from supply chains to marketing budgets.

The Foundation: From Business Problem to Optimization Model

At its core, prescriptive analytics relies on building a mathematical model of a business situation. The most common and powerful technique is linear programming (LP). Formulating a linear program involves three critical steps. First, you must define the decision variables. These are the controllable quantities you want to optimize, such as the number of units to produce or the amount of money to allocate to a project. Second, you establish the objective function, a linear equation that calculates the outcome you want to maximize (e.g., profit) or minimize (e.g., cost). Third, and most crucial, is constraint identification. Constraints are the real-world limitations expressed as linear inequalities or equations, such as production capacity, budget ceilings, or minimum quality requirements.

For example, imagine a company makes two products. The decision variables ($x_1$ and $x_2$) represent the production volume of each. The objective is to maximize profit: $Z=45x_1+60x_2$. However, constraints limit this: machine time ($2x_1+4x_2\le 100$ hours), labor ($3x_1+2x_2\le 120$ hours), and a demand limit for product 1 ($x_1\le 30$ units). The formulation process forces you to quantify the business problem precisely, which is often an illuminating exercise in itself.

Solving Models: From Excel Solver to Sensitivity Insights

Once formulated, you need a method to find the optimal solution. For linear programs, the simplex algorithm is the classic, efficient method for exploring the feasible region defined by your constraints to find the best vertex point. In practice, business analysts rarely code this algorithm from scratch. Instead, they use tools like solver implementation in Excel, which provides an accessible interface for defining variables, the objective function, and constraints before calculating the optimal solution. Understanding how to properly set up a model in Solver—including selecting the Simplex LP solving method—is a key practical skill.

After obtaining the optimal solution, the analysis is not complete. You must perform sensitivity analysis, which examines how sensitive the optimal solution is to changes in the model's coefficients. Two key reports are generated: the Shadow Price and the Allowable Increase/Decrease. The Shadow Price tells you how much the objective function's value (e.g., total profit) would improve if you could relax a constraint by one unit. This directly identifies bottlenecks and quantifies the value of acquiring more resources. The Allowable Increase/Decrease shows the range for which the current solution's variables remain optimal if an objective function coefficient changes, providing crucial guidance for pricing and cost negotiations.

Advanced Models: Integer Programming and Network Flows

Not all business decisions involve divisible quantities. When decisions are yes/no, or you must select whole numbers of projects or machines, you need integer programming. This is an extension of LP where some or all decision variables are restricted to integer values. A special case is binary programming, where variables are 0 or 1, used for capital budgeting, facility location, and project selection problems. While more computationally intensive, integer programming is essential for modeling discrete decisions realistically.

Two classic and highly applicable types of linear programs are the transportation problem and the assignment problem. The transportation problem minimizes the cost of shipping goods from multiple sources (e.g., factories) to multiple destinations (e.g., warehouses), given supply and demand constraints. The assignment problem, a special case, optimally assigns agents (e.g., employees, machines) to tasks to minimize time or cost, where each agent is assigned to exactly one task. These models have efficient, specialized solution methods and are foundational for supply chain and logistics optimization.

Bringing It All Together: Scenario Optimization and Decision Support

The ultimate goal is to build robust, decision-support systems. This involves scenario optimization, where you run your model under different "what-if" assumptions—fluctuating demand, volatile raw material costs, or new regulatory constraints. By analyzing the optimal solutions across these scenarios, you can develop flexible strategies and identify decisions that perform well under a range of possible futures. This moves prescriptive analytics from recommending a single, fragile "best" answer to providing a resilient portfolio of actionable recommendations, empowering managers to make data-driven business decisions with greater confidence.

Common Pitfalls

1. Ignoring Hidden Constraints: The most common error is failing to identify all relevant constraints, leading to an optimal solution that is mathematically correct but practically impossible. Correction: Engage with stakeholders from operations, finance, and logistics in the model formulation stage to uncover all limitations.
2. Misinterpreting Integer Requirements: Using a standard LP solver for a problem that inherently requires whole-number solutions (like building 2.5 factories) yields unrealistic results. Correction: Recognize the discrete nature of the decision and use the appropriate integer or binary programming setting in your solver.
3. Overlooking Sensitivity Analysis: Taking the initial optimal solution as a final answer without testing its robustness is a major risk. Correction: Always generate and interpret the sensitivity report. Use the shadow price to identify valuable resource investments and the allowable ranges to understand the stability of your recommendation.
4. Fixing Inputs as Static: Business environments are dynamic. Building a model once and never updating it renders it obsolete. Correction: Treat optimization models as living tools. Implement a process for regularly updating input data (costs, capacities) and re-running scenario optimization to validate or adjust the prescribed course of action.

Summary

• Prescriptive analytics uses optimization models to recommend the best course of action, moving beyond insight to direct guidance.
• The core technique is linear programming, which requires precise formulation of decision variables, an objective function, and constraint identification to model a business problem mathematically.
• Tools like Excel Solver are used for solver implementation, but post-solution sensitivity analysis (shadow price, allowable ranges) is critical for understanding the solution's robustness and value.
• For decisions involving whole units or yes/no choices, integer programming (and binary programming) is necessary.
• Specialized models like the transportation problem and assignment problem provide efficient frameworks for common logistical and allocation challenges.
• Effective application involves scenario optimization to stress-test recommendations under various assumptions, building resilient, data-driven business decisions.