Machine learning trials herald a new age of predictive accuracy for financial service firms
The results of a year-long machine learning trial conducted by Callcredit Information Group point to potentially significant positive benefits in predictive accuracy across a range of established data models.
The study looked at a number of different scenarios, ranging from identifying potentially fraudulent applications through to the accuracy of underwriting decisions in predicting a customer’s propensity to pay and bad debt across a portfolio.
The results of the study were very encouraging and point to potential significant financial benefits for adopters of the technology in the credit, fraud and insurance arenas. In one modelled scenario, the level of default in a portfolio of 60,000 credit cards was reduced significantly, resulting in a 10% reduction in overall bad debt. If used with other elements of the customer lifecycle, potential machine learning generated benefits could be even greater.
Mark Davison, Chief Data Officer, Callcredit Information Group, explained: “With many of our clients operating in exceptionally competitive markets we’re looking to build the next generation of predictive tools that will give them greater certainty and a competitive advantage, whilst also enabling them to provide a better service to their customers.
“At its heart, machine learning is a simple concept, it is a branch of artificial intelligence which enables computers to use an algorithm to analyse large amounts of data and discover new relationships or hidden patterns. Once these patterns have been found, they can be used to make predictions and solve a range of data-related problems.”
About the study:
The study looked to identify the best algorithm to solve a number of real-life problems and combined a set of business logic tools and a cloud based solution to develop a more accurate predictive model.
The trial used Callcredit’s own scoring data across a representative range of datasets, and deployed a Microsoft Azure ML solution. The research looked at a wide range of techniques and approaches and demonstrated a significant uplift by fitting a machine learning model over and above traditional scorecard approaches.
The solutions used mean users can deploy bespoke, specialist or value-add predictive models and analytics quickly and easily, generating significant commercial and competitive benefits.
Mark Davison continues: “Our trials are significant because they have looked at a selection of real-world issues. We have been able to compare existing results with those our new machine learning approaches would have generated and have been able to show some potential substantial benefits.
“Importantly, the solution we used recognises any migration to new technology within an organisation is often gradual. We tested a solution that would allow businesses to employ champion challenger models that could be run from the cloud, or enable them to swiftly retrain or recalibrate scorecards when necessary.”
Mark Smith, Senior Director, Microsoft, said: “Microsoft Azure provides Callcredit with a solution that allows it to radically speed up its deployment of predictive models and scorecards and to host, manage and publish web services in an environment that easily integrates into an existing infrastructure.
“Azure allows data science teams to publish models into an incredibly robust, scalable and fully secure cloud based infrastructure, providing faster and more flexible ways to deliver services.
“We’ve seen a trend where forward looking organisations are looking to use machine learning techniques to make smarter decisions with their data, Callcredit is one organisation exploring this revolution.”
Next steps and further reading:
Encouraged by the strong results of the trial, Callcredit is now testing the scalability of the new methodology in a range of use cases for a number of clients.
Aligned to its ongoing development of further machine learning solutions and tools, Callcredit has also launched its first in a series of white papers (Credit, Fraud and Risk in the Age of Machines), articles and guides on the topic. These are designed to help businesses better understand the opportunities machine learning represents and will act as a useful educational tool. The series will consider both current and potential future trends and applications.