by Brad Carr, Senior Director, Digital Finance, IIF

Industry View from

How machine learning is powering a financial services revolution

The promise of machine learning brings new opportunities and new questions to many facets of life, from autonomous vehicles to the media and marketing algorithms that present content to us online.

 

These technologies are now gaining momentum in the financial services sector, most prominently in areas of application that include analysing credit risk and detecting illegal activity such as money laundering and fraud.

 

At the Institute of International Finance (IIF), we’ve analysed our member financial institutions’ applications of machine learning through a series of surveys. Focused on machine learning applications in credit risk, surveys in March 2018 and August 2019 provide insights into the continuing evolution and progress of techniques such as gradient boosting and random forests.

 

The latest survey shows a sharp increase in the number of banks running pilot projects with these techniques, up from 20 per cent to 45 per cent. Although there has only been a modest increase in the number of banks using machine learning in production (up from 38 per cent to 42 per cent), the sophistication of these models has increased markedly.

 

Equally significant is the progress in the breadth of application across customer segments. In 2018, machine learning was primarily used for credit decisioning in retail portfolios, and with some other applications in credit monitoring in the wholesale and large corporate segments. This represented something of a bimodal distribution, where banks had large pools of existing structured data on retail customers, while there are new sources of unstructured data (such as external news services and supply-chain data) available for large corporates, where natural language processing can be applied.

 

But while there hadn’t previously been activity in the segments in between, 2019 has seen a sharp increase in usage for small and medium-sized enterprises (SME) portfolios, up from 8 per cent to 40 per cent of banks (see Figure 1).

Figure 1: Application of machine learning by portfolio type

More broadly across all customer segments, credit scoring and decisioning remains the most prominent area of application, but we’ve also seen significant growth in credit monitoring and the early warning signals for deteriorating credits, up from 13 per cent to 57 per cent of respondents, including the 25 per cent of surveyed firms that are using this in production. One notable initiative is at Scotiabank, where machine learning is used to identify credit card customers that may have trouble making their next payment. This is then used as the basis for proactively approaching those customers and offering them alternate arrangements, a move that has reduced arrears by 10 per cent.

 

The benefits (both expected and realised) of machine learning have been stable across years, including improved model accuracy, overcoming data deficiencies and inconsistencies, and discovery of new risk segments or patterns. However, banks’ perceptions of the key challenges in implementation have evolved considerably, encountering and identifying more challenges as their knowledge and familiarity with the technology has increased.

 

While data management (specifically bringing data from disparate sources into a single data lake), IT infrastructure and competition for the necessary human talent all remain challenges to completing a successful implementation, there is increased emphasis on supervisors’ understanding and consent to use new processes.

 

This reflects the fact that while banks have been becoming more mature with the technology, so have regulators, and so the nature of their scrutiny has matured with that. While the added scrutiny may intensify the challenges that banks could face in innovating, the fact that supervisors are increasingly able to ask the right questions is welcome and is beneficial for the future of safe innovation within the broader ecosystem.

 

Given its power and potential impact, machine learning requires a collaborative effort between the industry and the supervisory community to ensure that it protects customers without stifling its adoption or stalling innovation in the financial sector. Pilot projects that explore and test new innovations warrant encouragement from policymakers and supervisors, and it is heartening that the 2019 survey shows both the dramatic expansion of such pilots, and significant engagement between banks and their supervisors.


For more information please see the IIF Machine learning in Credit Risk 2nd Edition report here. 

Related articles

What's next?

Get our latest features in your inbox

Join our community of business leaders