PRESCRIPTIVE ANALYTICS: OPTIMIZING
DATA-DRIVEN DECISIONS
What is prescriptive analytics
What is Prescriptive Analytics?
Prescriptive analytics is the branch of data analytics focused on suggesting
actions you can take to affect desired outcomes. It builds on descriptive
analytics (what happened) and predictive analytics (what could happen),
but it goes a step further by prescribing specific decisions and strategies.
Key Features
• Optimization: It uses mathematical models and optimization techniques
to recommend the best course of action given certain constraints.
• Simulation: Often incorporates simulation models to understand how
different actions might play out under varying scenarios.
• Decision Making: Helps decision-makers choose among various options
based on data-driven insights.
• Uncertainty Handling: Prescriptive analytics can take uncertainty and
risk into account to provide robust recommendations.
We will focus on the first two points, although there will be elements of all four
permeating through the course.
Techniques and Tools
• Optimization Models: Linear programming, integer programming,
mixed-integer programming.
• Simulation: Monte Carlo simulations and scenario analysis.
• Technology: Common tools include Excel Solver, IBM CPLEX, @Risk,
Gurobi, and Python libraries like PuLP and SciPy.
Applications
• Supply Chain Management: Optimize inventory levels, transportation
routes, and production schedules.
• Healthcare: Develop personalized treatment plans and optimize resource
allocation.
• Finance: Allocate portfolios, manage risks, and optimize pricing strategies.
Overview of the cases we will cover
1
Resource Allocation (Excel Simulation
Decision Models: Solver) (@Risk)
Case #1: Red Brand Canners Dynatron
Case #2: XMicron
Case-based Report #1 Report #2
project:
Some real optimization use cases
• Bank branch network in Middle East
• Pricing for a large supermarket chain in Asia
• Inventory management at Amazon
• Resource allocation in Novartis
Why does “decision science” exist at all?
• Multiple Objectives: Maximize potential upside, minimize risk
• Different Alternatives: Infinite number of potential investments
• Sensitivity: How sensitive is the proposed solution to changes in exogenous
factors?
• Uncertainty of Key Elements: Volatile economy, covid pandemic, etc.
What is the value of the techniques we will cover?
• Objective: Reduce risk/uncertainty in decision making using available
information.
• Prescriptive vs. Descriptive: Not only explains why things happen but
also suggests solutions/improvements.
• More than a solution: Provides insights into situation, uncertainty,
objectives, and trade-offs.
• Adaptability: Easy to update when conditions change, accounts for
uncertainty.
• Support for Decision Maker: The model yields recommended actions
but should not replace the decision maker.
• Improving decision-making skills using decision models and technology.
“All models are wrong, but some are useful.”
— George E. P. Box
2
Optimization: A mathematical formulation
1. Single Variable Optimization
For a function f (x), the goal of optimization is to find the point(s) where f (x)
reaches a maximum or minimum. The critical points are found by solving:
f ′ (x) = 0
Steps:
1. Differentiate the function f (x).
2. Solve f ′ (x) = 0 for x.
3. Check the second derivative f ′′ (x):
• If f ′′ (x) > 0, the critical point is a local minimum.
• If f ′′ (x) < 0, the critical point is a local maximum.
2. Linear Programming
Linear programming (LP) is used to optimize a linear objective function subject
to a set of linear equality and inequality constraints for multiple variables.
General Form:
Maximize:
Z = C1 X1 + C2 X2 + · · · + Cn Xn
Subject to:
a11 X1 + a12 X2 + · · · + a1n Xn = b1
a21 X1 + a22 X2 + · · · + a2n Xn = b2
..
.
am1 X1 + am2 X2 + · · · + amn Xn = bm
Where:
• X1 , X2 , . . ., Xn ≥ 0
• b1 , b2 , . . ., bm ≥ 0
3
Key elements:
• Decision variables: Variables of the problem that lead to an optimal
solution (e.g., how much of each product should we produce, how much
should we invest. . . )
• The objective function: Maximize profit, minimize cost, etc.
• Risks/Constraints: Budget limitations, inaccurate forecasting, increased
uncertainty with more risk/constraints.
Steps:
1. Find the decision variables that define the problem.
2. Define the objective function (e.g., maximize profit, minimize cost, etc.)
3. Set up the constraints.
4. Use algorithms like the Simplex Method or Interior Point Methods
to find the optimal values of X1 , X2 , . . ., Xn .
