As each operational decision made in a Digital Decisioning system contributes to the performance of your business, each decision should be done in support of and in alignment with your broader business strategy. Each decision can be seen as an instantiation of your strategy, and, collectively, they determine your operational costs, business risk, customer loyalty, and financial return.
While the term “optimization” is used generally to describe the notion of improving decisions through analysis and tuning, there is one technology that provides a very rigorous and analytical approach. Also known as “Operations Research” and “Management Science,” mathematical optimization is applied to a specific problem by modeling the business objectives and constraints, instantiating the model with the actual data, and solving for the mathematically optimal or near-optimal decisions.
Mathematical optimization is a very powerful technique to apply to improve the quality of your operational decisions. It is model-driven and data-driven. By managing how you use shared and limited resources across decisions, it allows you to align your operational decisions with the objectives of your overall strategy. In contrast to a predictive analytic model, an optimization model is not something that gets generated. It is something that you write to represent the business problem at hand and use to “solve” the model to arrive at a set of actions.
Mathematical optimization can be applied operationally to a single transaction. It can be used to find feasible configurations of complex products or services, for instance, or to maximize the fit of a product or service to customer’s requirements. It can also be used tactically to optimize across a set of related micro-decisions that share a limited resource, incur a cost that needs to be used where most beneficial, or impose a risk that needs to be contained globally. Optimization can also be used to tune the thresholds or parameters of business rules in an existing Decision Service.
An Optimization System has a number of components:
- A Solver: This is a piece of software that uses a variety of mathematical approaches to final an optimal or best solution to a defined problem. Such problems generally involve an objective function that defines an outcome to be maximized (revenue, for instance) as well as a series of constraints on the variables used in this function. These constraints can be hard or soft. Hard constraints cannot be broken, while the solver will attempt to minimize the number of soft constraints being broken.
- Modeling language: A language is designed to allow the effective specification of the problem that is going to be submitted to the solver or optimization engine. Such a language supports the definition of variables and their allowed ranges, the specification of constraints in various ways, and the definition of the objective function.
- Design tools: A workbench for an optimization model crafts, tests, and tunes optimization models. The workbench often includes a set of tools to allow the model to be visualized so that relationships between elements can be explored and understood, along with tools to visualize results and performance and to compare various approaches to optimization.
- Interactive user interface: The interface supports business analysts in the creation of scenarios and the adjustment of model parameters. It allows the comparison of solutions and their business results.