Decision modeling practitioners frequently face situations when a decision model has to analyze multiple alternative decisions to come up with an optimal one. Traditional decision modeling approaches mainly rely on business rules and/or scorecards to determine one “good” decision and in most cases do not backtrack to consider potentially better decisions. The reason is simple: even for relatively small decision models with 20-40 decision variables it is physically impossible to manually create rules that cover all (!) possible combinations of decision variables and to find a decision that minimizes a total cost or maximizes resource utilization.
In this presentation we describe how “Decision Optimization” utilizes off-the-shelf optimization tools to help a business user to create smarter decision models that are capable to find decisions that minimize/maximize certain optimization objectives. Instead of trying to specify all possible rules, a decision model creator uses decision tables to represent only major business constraints and relationships between different decision variables. An optimization objective may be represented as a special decision variable defined on other key variables. A decision optimization component does the rest of work by automatically considering multiple alternatives and selecting the best one within a time limit defined by a user. Contrary to traditional rule engines, the optimization component is specifically designed to solve constraint satisfaction and optimization problems using proven optimization techniques such as constraint and/or linear programming. We will demonstrate decision optimization at work using two real-world use cases.