Methodical application of predictive analytics in businesses key to value

Business analytics systems aim to reduce inefficiencies in companies through the constant evaluation and modelling of data received from operations. However, the correct implementation is necessary to ensure the success of predictive analytics systems, says predictive analytics company Olrac SPS South Africa owner Dr Mike Bergh. He also emphasised that although business skill is widely available, the combination of strong technical/analytical skills combined with an in-depth understanding of business requirements is key.

Analytics systems are mainly deployed at particular points of business processes where predictions are required, he says.

"The models used by different analytics systems do not normally present a problem. However, a typical point of failure is when the business requirements are not properly understood before initiating the subsequent implementation steps.

"Further, the analysts who develop the models need to be involved in the specification of the business requirements, with sufficient feedback to ensure that what is being done meets business requirements," Bergh says.

Olrac SPS subscribes to the cross-industry standard process for data mining, which advocates a structured approach to the development and deployment of predictive analytics systems. The steps of the standard include business understanding, data understanding, data preparation, modelling and evaluation and deployment, he explains.

Data preparation is one of the most time-consuming aspects of implementing predictive analytics systems. The quality of the data determines the predictive power of the models. Any company that intends to use predictive analytics should ensure that a high quality of legacy data is available, Bergh says.

Meanwhile, about 80% of information exists in an unstructured format as text. Most modern data-mining tools offer the ability to convert unstructured text into structured quantitative data, he says.

"For example, comments recorded by call centre operators can be put through the text-mining component of a data-mining program to convert the data into a structured format that can be used in a predictive analytics model. This conversion and use of unstructured data often enhances the predictive power of the models that are developed."

However, other industries will have a greater focus on operational data. For example, hundreds of sensors monitor myriad aspects in the performance and status of large turbines and generators involved in electricity generation at a power plant, he says.

"In the context of a plant, predictive analytics can be used most effectively to predict the need for a maintenance shutdown or flag an impending equipment failure. In such cases, the data represents the readout of a large number of sensors at predefined time intervals, be it a few seconds, minutes or hourly intervals, rather than client details."

"Further, predictive analytics can form part of a large inventory control solution, such as, for example, auto-teller machines (ATM). A predictive model would be used to forecast demand at each ATM and would be part of a larger model to reduce the total amount of cash required at different ATMs."

Improving the logistics of replenishment, including frequency and triggers, can be achieved using optimisation software. Modern optimisation software can monitor and manage thousands of decision variables and constraints. The capabilities provided by modern systems and software highlight the benefits of improved computer software, concludes Bergh.