Optimising complex business problems requires sophisticated technology. Recent years have witnessed major advances in the speed of optimisation algorithms and in the complexity of problem that can be addressed. The net result is the proliferating use of optimisation technologies to address everything from marketing campaign optimisation to how many business class seats should be allocated on individual flights.
There are several well defined types of problem that optimisation techniques can address – and some they can’t. The earliest and often easiest form of optimisation assumed that variables and objectives were related to each other in a linear manner. If a resource usage is doubled, so is its cost. While there are some problems that are well served by this model (the use of material components in a mix for example), many are not. To cater for more complex optimisation problems, non-linear relationships have been accommodated.
A good example here is a price/demand curve where demand drops off rapidly as price exceeds a certain threshold, and increases exponentially as price drops below a critical level. The solution of non-linear optimisation problems is much more complex than linear problems, but contemporary tools with good user interfaces help keep such problems manageable. Other problems require that variables can only take on integer values (we can’t have 2.5 airplanes for example). Another class of problem makes use of network programming, where the aim is to minimise some function of the network. A good example here is minimising the cost of transport as a given number of trucks ship goods to a network of stores.
Other techniques are also finding their way into prescriptive analytics, in addition to the optimisation techniques mentioned above. Queuing problems are common in business and optimisation techniques are used to address problems from traffic flow through to minimising check-out queues in stores. Simulation is also used to model the performance of business systems and is a large domain in its own right. It is very often the case that the ‘best’ solution to various business problems simply cannot be found, and so looking for a good solution becomes necessary, and this is where both analyst and business managers need to really understand the problem they are attempting to solve.
Stochastic optimisation takes prescriptive analytics into a realm where many uncertainties in business can be accommodated. Employee attendance, future sales, the response to marketing campaigns, wastage and hundreds of other variables are inherently uncertain in nature. The variables can be treated as random in many ways, with limits on how much they can vary. The stochastic optimisation algorithms will find the best, or at least a good, solution for the most likely outcomes where uncertainty is present.
Such is the advanced nature of some prescriptive analytics tools and solutions that near real-time optimisation can occur to accommodate changing business conditions. For the very largest optimisation problems this still is not possible, but the frontier is being pushed forward rapidly and in volatile markets real-time optimisation can, and often does deliver significant benefits.
Prescriptive analytics technologies will advance rapidly over the next four to five years with new entrants and new capabilities. As always integration is largely the determiner of how successfully prescriptive analytics can be used in a live production environment. Building prescriptive models is one thing, using them in a production environment requires extensive integration capabilities and good management and control tools.
The next article in this series is: Prescriptive Analytics Methods