Australia has one of the highest rates of personal debt in the world. Estimates indicate that the average Australian owns at least 1 credit card and that roughly 57% of our $2 trillion personal debt is attributable to mortgages. These statistics lend themselves to the assumption that majority of Australians have at some point in their life experienced the process of applying for a line of credit.
Typically, the process of applying for a loan consists of a lengthy submission outlining personal and financial information, providing identification documentation (driver’s licence or passport), gas bills, water bills and months’ worth of pay slips and then finally a little prayer as the application is submitted for credit assessment. While the application process can be daunting for customers, it is the credit assessment that ultimately mystifies customers and frontline staff alike.
Traditionally the credit approval stage in the lending lifecycle is often the most disjointed and elusive step for bank customers, and sometimes even the lenders themselves. Financial institutions have seen many internal and external forces that are disrupting the traditional role of the bank as well as the competitive landscape, the very products they offer and their own organisational structure. As banks embrace robotics, digital frameworks and new payments platforms, the spotlight of change is generally directed towards the optimisation of the customer journey and acquisition, the credit assessment stage should be no exception.
The Dynamic Decision
The credit assessment doesn’t need to be a black box and there are benefits to making it a fluid and integral part of the customer journey. Engaging a decision engine (data-driven, rules based engine) at key points in the application process, it enables a flexible flow of credit and policy insights which allows the lender to engage in a powerful customer conversation. The decision engine can also be used to reach the customer with valuable insights at the moment which influences them the most.
“Dynamic decision making” offers customer and lender insights into the health of the application (i.e. serviceably, security, evolving loan to value ratio, policy breaches, concentration risks, etc.) at any point during the application process (subject to data capture of course). Decisions are indicative (i.e. draft, not final), which enables the customer to gradually build the best case for their submission. By the end of the application capture, the customer, lender and bank are confident with the level of success of the loan.
Using Key Data Sources
With the spotlight on responsible lending, data accuracy and quality has never been more crucial for lending institutions.
The age of relying on photocopies of payslips and customer inputs are over. The gathering of customer information for risk assessment is transforming from passive information retrieval to proactive information gathering.
Lending Institutions should leverage the wealth of internal and external data sources to accurately tell the customer’s story and determine the risk potential.
Pooling information directly from the source can improve the overall data quality, operational efficiency and in many cases where these data sources are directly exposed to a decision engine it can enable automation of credit assessment and key processes, thereby dramatically reducing time to a decision. 3…2…1… Yes!
- Credit Bureaus: Comprehensive Credit Reporting (CCR) provides a more complete picture of consumers and their credit-related behaviour. CCR supports responsible lending allowing banks to identify credit stress or over-commitment at a much earlier stage.
- Valuation Data: Sourcing historical and current valuation data coupled with the use of leading market analytics and indicators can help banks predict future trends in property markets.
- Rental Valuations: Allows for automatic verification of a customer’s rental income.
- Internal Concentration Risk Reporting: Automatically determines the banks’ exposure risks in any given property development.
- Open Banking: Utilising data from a customer’s other financial institutions to gain greater insights and more reliable customer data, which enables banks to automatic verify the customers financial position – No more pay slips!
- Utilities Open Data: With the use of API’s banks can auto verify a customer address, time at address, regularity/consistency of bill payments all from retrieving data from gas, water or electricity companies (with the customer consent of course).
- Biometric Data (i.e. Passport Data, Drivers Licence): By retrieving the biometric data from Traffic Authorities or Passport Office for example, banks can automatically verify the identity of a customer simply by comparing the biometric data to a ‘selfie’.
- Social/Behavioural Data: Overseas lending institutions have already begun to use Facebook and other social media to verify the identity and reliability of potential customers. This data is especially useful if the borrower does not have a credit footprint, for example students or graduates. A score is determined based on a number of factors, for example: Number of friends/followers, quality of contacts – which could indicate higher earning potential, profile of employment, location, time of day and use of the website.
Banks can rest easier as decisions are made with data found directly from the source which is generally more accurate, while their customers enjoy a more intuitive and personalised process.
Machine Learning and Credit Risk
Machine learning is not a new concept to lending. In fact one of the first uses of machine learning was within risk modelling to analyse data and forecast default risks.
Traditional lending scorecards and risk assessments are imperfect, requiring human interpretation and assumptions about risk. However as the use of artificial intelligence continues to develop within lending the need for human judgment and manual workflow is greatly rationalised.
Using algorithms to analyse large amounts of both financial and non-financial data adds a new layer of granularity and deeper scrutiny to enhance analytical competencies in risk management and compliance. This helps institutions make more informed decisions as now they are able to learn from the “exception cases”, to push past the black and white and into the grey areas of thought and rationale.
As the Australian banks push forward with increasing the presence of AI in their everyday operations, there are considerable learnings which can be taken from the European banks who have leveraged machine learning and have found with the adoption of the General Data Protection Regulation (GDPR) many regulations regarding algorithm decision-making have been enforced. Similar regulations are likely follow for Australia.
GDPR states that decisions regarding individuals must be non-discriminatory and biometric information should be solely used for the purpose of identifying the customer. However, it is the customers “right to an explanation” which should be carefully considered when introducing machine learning into decision making. Institutions must have the ability to explain how they arrived at the decision and therefore require a deep understanding of the data sets and algorithms used to reach each result.
Of course this scarcely scratches the surface when it comes to applying machine learning within decision making. In part II of this blog we will deep dive into what we’ve learnt about the application of machine learning to decision making, and tips for how to ensure success.