Structured Decision Modeling Leads to Better Strategic Outcomes
What is Decision Analysis?
Decision analysis is a collection of logical and methodical approaches to decision-making, including modeling for conditions of uncertainty, studies of decision-making behavior, and techniques such as decision trees and expert systems, according to the Decision Analysis Society of the Institute for Operations Research and Management Science (INFORMS). Investopedia is more succinct, defining it as “a systematic, quantitative, and visual approach to making strategic decisions.”
Decision Modeling Across Industries
Decision analysis was first introduced as an idea by Robert Howard, a professor of Management Science and Engineering at Stanford University. Today the concept is applied within a wide range of fields, from business and engineering to healthcare and public policy. Decision analysis is applied to many different types of decisions, including capital investments, strategic direction, bidding, litigation strategy, and healthcare planning. Examples include siting decisions on where to construct a plant, whether to sell or develop a patent or other IP, and building a product or system versus buying it. The goal of decision analysis is to provide decision-makers with alternatives when attempting to achieve particular objectives, while also accounting for uncertainties. It also provides measures of how successful objectives will be if various discrete outcomes occur.
Methods for Performing Decision Analysis
Methods for decision analysis often incorporate qualitative and quantitative factors, such as management opinion, psychology, and economics. Tools and methods vary, and include techniques such as decision trees, influence diagrams, optimization, gap analysis, analytical hierarchy processes (AHP), and expert systems. Both AHP and expert systems methods formalize the criteria by which a human expert would judge different alternatives, and create series of if/then rules. Once the systems are defined with input from experts, automated decision-making is enabled for future occurrences of the same or similar situations.
Decision trees are available for any analyst at the desktop level, directly within Microsoft Excel. This provides maximum modeling flexibility with a minimal learning curve.
A decision tree showing multiple decision options.
Influence diagrams consist of nodes representing uncertainty (probabilities), decisions, and values (payoffs). The nodes are connected by arcs, which are functional (ending in a value node), conditional (ending in an uncertainty node), or informational (ending in a decision node). Unlike decision trees, influence diagrams are not sequentially structured. Influence diagrams are meant to show (see below):
- The available alternatives in a situation (as depicted by decision nodes and incoming information arcs)
- The available information (as depicted by uncertainty nodes and incoming conditional arcs)
- Preferences (as depicted by value nodes and incoming functional arcs)
Like decision trees, influence diagrams are available for any analyst at the desktop level, directly within Microsoft Excel.
Simple influence diagram.
Though computationally intensive, optimization tools are available at the desktop level for any Excel user, where many decision-makers share and construct decision models.
An optimization improving on different solutions.
Decision Analysis and Risk Analysis
Some practitioners view decision analysis as a form of risk analysis, while others view it vice versa. Either way, there is no doubt the two disciplines are connected, and both improve overall decision-making.
One way that risk analysis can be used to enhance decision analysis is through the application of stochastic risk analysis, or Monte Carlo simulation, to decision analysis. In most cases, decision analyses only concern themselves with multiple discrete outcomes. Typical decision analysis methods are unable to account for the real-life continuous nature of what are effectively infinite outcomes. This is an area to which Monte Carlo is well-suited.
Monte Carlo Simulation and Decision Trees
A Monte Carlo simulation run on a decision tree.
One way to address the issue is to apply Monte Carlo simulation to decision trees. Use probability distributions to represent the continuous range of possible chance outcomes at any given chance node. You can also apply distributions to final outcomes, or end nodes. Thus, a tree that formerly contained a half dozen discrete end nodes may be consolidated into a single end node represented by a probability distribution. This not only makes the tree more accurate, but more manageable as well. A simulation can then be run, providing a probabilistic view of what various decision strategies will lead to.
Monte Carlo Simulation and Optimization
A Monte Carlo simulation comparing the optimized solution against the original, pre-optimization model.
Traditional optimization, like decision trees, makes assumptions about the certainty of factors that are, in reality, uncertain. For example, an optimization model may include fixed values for unknown variables such as how long a task takes, what the return is on an asset class, or what the demand for a product will be. As a result, the “solution” (or recommended allocation) fails to take into account variability. The solution may show a maximized resulting return, for example, but that return may have a lot of uncertain around it.
To address this problem, Monte Carlo simulation can be run in conjunction with optimization. Replace fixed values for unknown variables with probability distributions. Then, for every allocation strategy the optimization algorithm tries, run a Monte Carlo and record not just the expected result (usually represented by the mean), but the variability around it (often represented by standard deviation). This enables you to set more intentional objectives as well, such as maximizing a return subject to keeping standard deviation within acceptable bounds.
Decision Analysis with Palisade
Palisade’s PrecisionTree software puts powerful decision analysis at the fingertips of any Excel user. PrecisionTree makes it easy to create, analyze, and share decision trees and influence diagrams with others, all from the familiar Excel environment. Furthermore, PrecisionTree is designed to work with Palisade’s @RISK and TopRank products, which add Monte Carlo simulation and sensitivity analysis to decision tree models.
Palisade’s Evolver and RISKOptimizer tools bring robust optimization techniques to any Excel spreadsheet. RISKOptimizer further adds Monte Carlo simulation, accounting for uncertain factors and a broad range of outcomes.
All products are available in the DecisionTools Suite. These combined analyses allow the creation of the most accurate decision analysis models anywhere, while maintaining point-and-click ease-of-use.