Help is on the line

Is it taking literally what they say? Do you also factor in what you understand about them?

"All this means the system can now spot customers in need up to 30 to 40 days earlier than the previous process."

A century ago a bank was like a corner shop. We knew the manager and they knew the ins and outs of our finances. It was their job to listen.

As banks grew into the giant institutions they are today, what an institution knew about customers could take three to six months to process by observing trends in account balances.

If a customer was heading for trouble financially, it could be too late once the problem was detected.

Even up until a decade ago, data could be a month old before it was included in a customer assessment.

But the bank of today now has billions of data points available - almost in real time.

This has allowed ANZ to transform its predictive power by developing and testing a daily transaction score, developed in conjunction with US based analytics company FICO.

The solution is based on the use of intellectual property (IP) and technology that draws on decades of experience in transactional data analytics from FICO’s ground-breaking fraud solution FICO Falcon Fraud Manager.

It is a development which shows a future in which a responsible use of Artificial Intelligence isn’t a theoretical benefit, somewhere down the pipeline, but a tangible asset that is already helping customers today.

What is a transaction score?

A “transaction score” is calculated via the raft of data that can be fed into machine learning, giving a score to an account where there might be some early signs of financial stress.

This allows bankers to get in touch, by either SMS or telephone, to see if there’s anything the bank can do to help before things get too bad.

Our new customer score recognises we want to do more but we want to do it the right way. But the question we asked ourselves was, with a sea of data and so many options, how do we use the technology in an effective way? 

The first part was about having a sound theoretical grounding.

A key part of ANZ’s approach was starting from the bank’s larger strategy of having a goal to improve the financial wellbeing and sustainability of customers.

This derives from our embrace of the Australian Banking Association’s Banking Code of Practice which directs us to employ a range of practices that can identify common indicators of financial difficulty and reach out if customers are having trouble.

As ANZ’s Head of Financial Wellbeing Mohamed Khalil says it is a key Environmental, Social and Governance goal to have the work of financial wellbeing and behavioural science experts embedded into our innovation processes.

Nudge, Nudge

But how do you transform a vague industry standard into a reality?

The scoring program relies on pioneering behavioral economics work by Nobel prize awarded economist Richard Thaler.

Thaler proposed using design to help customers make smart decisions for their health or financial well-being – a technique dubbed “Nudge Theory”.

In the past, with fewer bits of information available, banks scored based on balances of accounts and ratios of payments to balances. Under the new transaction scoring system, ANZ looks at what makes up the balances and the internal dynamic: what is going on “underneath”.

Is cash going out? Is it being spent on shopping? What sort of shopping was it?

All this allows better predictions and better modeling.

It can pick up when a customer is suddenly spending excessively compared with savings and earnings. Or it can see when there is a high level of internal transfers between savings and transaction accounts. It can even reflect upon when a customer suddenly moves onto government payments.

All this means the system can now spot customers in need up to 30 to 40 days earlier than the previous process. It also means ANZ has increased by about 3000 per cent the data it is using to work out if customers can be helped.

As a part of establishing the system it oversaw a total of 5 billion transactions over a 36-month period. In the pilot stage the scoring system has assessed the last year of transactions and created a score for 7.7 million customers. This translates to processing of 17 million accounts and 15 million direct debit and credit card transactions a day.  

But these are not just theoretical victories, our testing shows the new scoring system is already proving a success. In the pilot stage about 40,000 SMS messages have been sent to retail and business customers who operate small to medium sized companies.

This is challenging old ways of thinking.

Customer scoring has traditionally focused on retail customers but at ANZ we consider all customers in the retail and commercial lending and nonlending space. We are working to increase the breadth of inclusiveness.

In another first, the score applies to customers who bank with ANZ but may not have a loan. So the bank can understand the customer’s needs and supply a credit rating even though lending has not taken place.

Eventually ANZ’s “nudges” will be conducted in the ANZ App. The score system has also been tested on communities affected by the NSW floods and proved successful in offering a helping hand to those who – through no fault of their own – were in danger of suddenly experiencing financial hardship.

The response was extremely positive as people recognised their bank had reacted quickly to unfolding events.

This is new technology strengthening existing relationships. So far, the testing of the score has improved accuracy and outcomes across risk bands by 8 to 18 per cent.

The work has been recognised as a world first and was in May this year awarded a FICO Decisions Pioneer Award at a ceremony in Hollywood, Florida.

“ANZ Bank has already put this strategy to good use during challenging times. By using transaction scoring combined with predictive analytics, they were able to identify customers affected by natural disasters, such as floods and fires, who might need debt management assistance,” FICO’s executive client partner in Australia Corey Smith said.

“This allowed them to offer support proactively and much sooner than they would have been able to previously.”

By intervening early, customers are helped to get their affairs in order before they spin out of control.

What does this mean for a customer who wants to use ANZ products?

It means credit card authorisation capabilities are improved. It assigns better limits to customers. It reduces manual treatment for customers sent to collections. And it identifies in a more exacting way the small number of higher-risk customers who need appropriate management.

Ultimately, it's all about empowering our customers to make the most of their financial lives and grow their businesses, while also managing risks effectively.

And while the techniques of what ANZ is doing are new, the theory is old. Prior to the 1990s, we transacted with a bank by speaking to the manager who would use transactional data - in ledgers - to decide whether you got a loan or not.

In the digital age, transactions have increased, customers have increased and behaviour has changed, making this a big data problem to solve. However, the process still stems from those bank manager conversations in which they “know” their customers by looking at their transactional behaviour.

Transaction scoring brings back that element of personal decisions while ensuring we are connected to customers in the moment-by-moment digital world.

Jason Humphrey is the Chief Risk Officer of ANZ Australia