The bank's Pega-based customer engagement engine handles inbound and outbound communication with its retail customers through all channels.

It is predominantly about service messaging, he said, such as customers updating their contact details, the bank sending birthday greetings, or ensuring that contact centre agents are aware of special circumstances such as natural disasters affecting particular customers.

The engine is also used to send reminders to help customers avoid bank fees, for example by alerting them to the need to pay the balance owing on their card in the next few days so they don't incur credit fees. This is part of a program that will save customers $400 million in fees this year, said Burton.

The system "works very well" from the bank's perspective, he said. While Burton was unable to provide any metrics around customer engagement and retention, he did say that it has improved the NPS (net promoter score) in areas including the CBA mobile app, NetBank (internet banking), and branches.

Looking ahead, CBA wants to increase the number of real-time data feeds to the engine, including transactions (iTWire heard a similar story from HSBC Australia). That would allow better engagement and personalisation, for example by preventing reminders being sent just after customers have paid their bills.

Other possible uses include warning customers when their projected account balance is insufficient to pay a predicted bill.

It is important to avoid "creepy" behaviour, but this can be achieved at least in part by ensuring you have the customer's consent as well as rightful purpose. "These rules keep getting stronger and stronger," Burton observed.

The bank also needs to be able to explain any decision by the system, and that includes understanding choosing not to take a particular action is as much as decision as choosing to do it.

But because these decisions are being made in real time, it is necessary to keep all the data as it was used in the decision process – merely knowing the customer's account balance at the end of the day might not be sufficient.

As the amount of data used in the process increases, the storage needed to meet legislative requirements could become impossibly large, which would mean Australian banks may have to quarantine some of their customer data from AI-based processes.

Notwithstanding that potential problem, Burton suggests CBA customers can expect their communications with the bank to be increasingly personalised based on things they do and don't do, and that this will extend to additional channels through the use of technologies such as voice and natural language recognition.

Disclosure: The writer attended PegaWorld 2019 as a guest of the company.