Optimisation sits at the heart of prescriptive analytics technologies, and specifically the computation of best resource usage given a set of constraints and objectives. Work planning problems represent a classic application, where work is allocated to limited human resources in a manner that meets constraints and optimises objectives.

While optimisation has been used for decades in many large corporations, the compute intensive processing has traditionally been associated with very long compute times – typically days and weeks. This limited the application of the technology. However advances made in the mathematical algorithms and more powerful hardware mean that optimisation can be applied to a much broader range of problems, and in some instance execute on a near real-time basis.

The three essential components in an optimisation problem are variables, constraints and objectives. In a work planning problem the variables would typically represent the number of hours work allocated to various people from a given list of tasks. The constraints would limit the way the allocation of resources could take place – no more than 20% of the personnel from any department can be engaged on a project for example. Finally the objectives state what we are trying to achieve. Often this is simply to minimise costs, or maximise profits – or both. However in the work planning problem we might be most interested in minimising the time a project takes. Each optimisation problem has its own set of variables, constraints and objectives and much of the work goes into specifying what these are.

Prescriptive analytics can be divided into two primary activities. The first involves optimisation when the input variables are known (a stock count, or balances in accounts for example). The problem here is simply to establish the best outcome given these variables along with associated constraints and given objectives. A second set of optimisation problems comes under the heading of stochastic optimisation, a suitably off-putting name which simply indicates there is uncertainty in the input data – next monthâ€™s sales for example. This more complex category of problems will attempt to find the best solution to a business optimisation problem for the most likely future situations. Obviously there is a strong link here with statistical modelling and other forms of predictive analytics, where probabilities are assigned to variables.

It is increasingly the case that prescriptive analytics is integrated with other systems. Optimisation has traditionally been an isolated activity, but today it can take inputs from business rules and predictive analytics processing, and benefits hugely from them. The business rules act as constraints (do not mail someone with an offer of a 5% discount when they have already been mailed a 10% discount – for example), and predictive analytics can provide inputs which predict variable values (the number of prospects likely to respond to a marketing campaign for example).

Prescriptive analytics is still relatively new (the term was first introduced about a decade ago) and only a handful of suppliers provide the integrated environment necessary to take advantage of outputs from other processes. However prescriptive analytics does complete the analytics picture – descriptive analytics (business intelligence) and predictive analytics say **what** has happened or will happen, while prescriptive analytics say** how** things should happen.

The next article in this series is: Prescriptive Analytics Methods