How CommBank is training its machine learning Customer Engagement Engine

The batch of customers that are responding to a simple notification are actually helping the bank train its 200-plus machine learning models.
Written by Asha Barbaschow, Contributor

The Commonwealth Bank of Australia (CBA) has built a Customer Engagement Engine it has touted as powering customer experience through the use of artificial intelligence (AI) and machine learning.

"[It] is an advanced system which combines AI, machine learning, and our customer data to continuously optimise and prioritise across all of the available messages, alerts, conversations, and communications we can have with our customer at any given time, across all of our channels," CBA chief analytics officer Andrew McMullan said.

"And all of that in real time."

Speaking with media at the launch of the bank's new app, McMullan said CBA runs over 200 machine learning models on top of 157 billion data points. He said these models are continuously "optimising, prioritising, and returning the next best personal message to our customers as they interact with us across all channels".

McMullan said the bank has invested a lot in automated machine learning capability.

"So maybe there's a message that we want to test with some of our customers, in one of the assets within the mobile app, we'll push that message live and in the background we'll switch on one of our advanced machine learning models," he explained.

"As our customers engage with that message, maybe a click through a 'yes, please', or 'no, thanks', the machine learning model in the background is learning from that to really more accurately predict which type of customers really enjoy experiences with that particular message."

When CBA is happy with the performance of the model, McMullan said it would switch that conversation on and add it to the system to become another message that will be shared with customers.

"As I navigate my way through the experiences in the mobile app, each of the assets has been personalised by making a call to the Customer Engagement Engine to say, 'What's the next best message to determine this particular communication?'"

See also: How to increase employee engagement using AI, machine learning, and other methods (TechRepublic)

Providing a scale of the system, McMullan said that each day the bank's applications make over 20 million calls to the Customer Engagement Engine to return personalised messages.

Over the next 12 months, the bank expects it will have the opportunity to deliver 3 billion personalised messages to its customers.

"One of my favourites is our Smart Alerts ... on credit cards. We're constantly monitoring our customer data and if a customer hasn't paid the minimum amount or paid down their credit card balance, three days out from being due, we will send them an alert,  make it really easy to click through to make a payment," McMullan said, offering an example of how the bank is using the Customer Engagement Engine.

"If you haven't done that and your payment is due in two days, another alert. One day, same day, even one day's grace. The last 12 months alone, we've built the ability to send over 20 million alerts to our customers to help them avoid fees and charges on credit cards."

Another example is how CBA is trying to help customers manage their bills.

McMullan said that by using natural language processing, the bank is attempting to better understand all the regular payments and bills that a customer has in order to identify any anomalies.  

He shared a personal anecdote where he was on a monthly subscription that had a price hike.

"You can see [from that] example there are hundreds of opportunities for us to just notify our customers and let them know that something has changed, and then they can decide what to do from there," he said.

The final example McMullan shared was helping the bank by using the system during the application process.

"Our customers begin an application and for many reasons they drop out. Maybe they haven't submitted the document, finished entering a specific field, or even signed the document. The Customer Engagement Engine is constantly scanning over that, identifying exactly what the customer needs to do next, and we'll reach out to the customer to let them know what it is that they need to do to continue the application process," he explained.

"We are determined to improve the financial wellbeing of our customers and  communities."

According to McMullan, for customers that have been receiving the personalised messages, the net promoter score is up to six points higher than it is from customers that don't receive messages.


Editorial standards