Predictive analytics is concerned with the analysis of historical data to discover patterns of behaviour which might be useful in the future. The classic example is credit checking, and trying to establish who will, and who won’t be a good credit risk in the future based on the analysis of historical data. Predictive analytics is quite different from business intelligence and reporting activities most organisations engage in. These tend to be called descriptive analytics, and as the name implies it is concerned simply with describing what has happened in the past and what might be happening now. Predictive analytics is concerned with the future.
To attempt predictive analytics an organisation needs relevant data. If we are trying to establish which customers might be the best candidates for a promotion then we clearly need good quality customer based data. A great deal of effort is usually expended in making sure the data is as accurate as possible, since predictive analytics conforms to the garbage in – garbage out paradigm as much as anything else. But assuming we have decent data the next step is to use an analytics platform to prepare data and analyse it. There are many algorithms used in predictive analytics and knowing which ones to use requires a good deal of experience. In any case the algorithms will usually identify patterns of behaviour and indicate which variables are important. The resulting models that are built need to be validated and checked by a domain expert, since analytics tools can be fooled by random coincidences.
Models deemed suitable can then be used to score new data. In the case of credit approval the predictive models will often produce a score where a threshold limit has to be reached to grant approval. In many industries it is absolutely essential that the inner workings of the model are understood – a person cannot be refused a loan simply because a piece of software says so – the reasons need to be understood by humans.
Predictive analytics is most widely used in customer related activities within many organisations, and can address issues such as churn rates, marketing, selling strategies and so on. But it is also used to identify when a machine may fail, or when a patient is likely to be readmitted into a hospital, or even when an employee might resign.
For ‘standard applications’ and particularly in sales and marketing activities there are many technology suppliers selling ready made solutions, and so the user does not need to understand how the technology works. In other applications however it is necessary to employ skilled data scientists and analysts to create a useable predictive model. This is especially true of big data (a meaningless term really – it’s just data) where special consideration needs t be given to the nature of the data.
These are early days for predictive analytics and the underlying technology, mainly derived from statistics and machine learning, is advancing quickly. It is already used to determine how goods are laid out in a store, which movies are recommended to a particular viewer, and who might be at risk of various diseases. It will become ubiquitous and affect every aspect of business and our lives.