Major bank exploring use of AI to assess loan serviceability

Speaking at the NVIDIA AI Conference in Sydney, ANZ Bank’s head of retail risk, Jason Humphrey, revealed that the major bank, in partnership with Nvidia and Monash University, has developed a proof of concept that demonstrates how applying artificial intelligence to transactional data can help lenders assess risk faster and more accurately.

The proof of concept used a neural network and credit card data from one million accounts to predict which customers were likely to default on their loans, with the head of retail noting that proactively identifying high-risk loans is key to limiting ANZ’s exposure, while reducing the need to maintain large asset reserves.

Mr Humphrey said that there are two key methods that have historically been used by lenders to assess risk: application scoring and behavioural scoring. The former combines information contained in applications with bank data and credit bureau data to measure the probability of a customer defaulting on their loan; the latter uses a lender’s master file data to predict the behaviour of a customer under various credit limit, interest rate and product scenarios.

While behavioural scoring generates better predictions than application scoring, with lenders able determine serviceability based on certain types and sequences of transactions, the head of retail noted that the method is limited by the availability, accuracy and volume of data, as well as the frequency at which a customer’s risk profile can be reassessed. Further, behavioural scoring usually compares the monthly closing balances of borrowers, with no further granularity.

He said that two customers could have the same balance at the end of the month, but one could be a riskier bet than the other based on the timing and nature of the transactions.

However, these problems are likely to diminish over time. Mr Humphrey indicated that ANZ is preparing to take advantage of the impending open banking regime, set to come into effect on 1 July 2019, which will allow customers to share their transactional data between authorised financial services providers.

“We receive, just in ANZ, over 10 million transactions a day, transactions that are important, in modelling, which we’ve never been able to touch until our partnership with Nvidia,” the head of retail said.

He added that over 12 months, the bank recorded 1.7 billion transactions from individual consumers and 300 million transactions from small business consumers.

“The opportunity to dig behind balances and transactions is very exciting,” Mr Humphrey said.

The head of retail spoke favourably about a company in Canada that uses the quality of products — such as the grade of motor oil a borrower purchases — as indicators of the customer’s ability to manage risks. This type of data is drawn from receipts.

He added that through daily behavioural scoring (as opposed to monthly scoring), lenders would be able to “see risk changing more dynamically”.

“The key to risk management, banking and finance is time,” Mr Humphrey said.

“The quicker you can resolve a problem, the quicker you can minimise any loss, inconvenience or degradation of customer experience.”

The revelation of the proof of concept comes as lenders tighten up their credit policies around income, expenses and benchmarking against the backdrop of an ongoing financial services royal commission, which highlighted many cases where non-qualifying loans were approved, only to end up in default.

Commonwealth Bank notified mortgage brokers that it would be looking at 11 categories of living expenses on loan applications, while Westpac updated its expense guidelines, requiring documentation at an “itemised and granular level” across 13 categories. ANZ followed suit, cautioning brokers to ensure that an “accurate reflection of the customer’s financial position is presented”.

While the banks are under regulatory pressure, at the other end, mortgage brokers have expressed frustration over the increasing interrogation of living expenses, saying that lenders are palming off responsibility to brokers. The Australian Securities and Investments Commission had also said that lenders shouldn’t offload blame to third parties when a loan is found to be in breach of responsible lending obligations.

A recent study by financial services comparison site Mozo found that there is a growing league of “mortgage prisoners” in Australia who are unable to negotiate or refinance their home loans due to tighter lending criteria around income and living expenses. These borrowers are also “vulnerable” to any interest rate hike that is set by the banks, as a result. 

[Related: Banks facing mortgage market ‘dilemma’]