The Commonwealth Bank of Australia is continuing down its generative AI path with the emerging technology now embedded into a number of customer-facing capabilities.
At a ‘Reimagining Banking’ briefing, initially intended to be held earlier this year, the bank showcased its technology, AI-powered features and its future thinking.
This included a 50 percent reduction in customer scam losses supported by AI-based checking technology, including NameCheck, CallerCheck and CustomerCheck.
CBA said it had also seen a 30 percent reduction in customer-reported frauds due to measures like generative AI-powered suspicious transaction alerts.
Its AI-powered app messaging led to a 40 percent drop in call volumes over the past financial year.
CBA CEO Matt Comyn said the results are a testament to the accuracy of the models and of their output, noting that it is using GPT-4.
“We’re also doing work with AWS, Anthropic, and directly with NVIDIA as well,” he said.
“If you use that [model technology] personally, you will see over the course of the year how much better the model responses are.”
He said CBA’s chatbot experience “just with the model improvements” has seen “a 20-percentage point uplift in terms of what’s possible in the new version that we’re bringing to market versus what we’ve had in place.”
Generative AI messaging
CBA’s group executive for retail banking services Angus Sullivan said there are now 12 million logins for its banking app per day, and 38 percent of all new trading CommSec accounts originate from the app.
He said that of the 20 million payment transactions a day processed by the bank, about 15,000 transactions “are disputed”.
“What we’ve done over the past 12 months is dramatically increase the volume of proactive alerting to customers around transactions to check if that transaction is indeed a proper and safe transaction,” Sullivan said.
“We’re now alerting customers to say, ‘Hey, can you just check this a transaction you’ve made?’”
This was driven in part by a “need to be able to have a real-time, fully digital, end-to-end experience for resolving the dispute”.
“Otherwise, we’re sending out a lot of alerts, we’re creating a lot of engagement with customers [but] they’re going to call us,” he said.
“That’s going to create an enormous call volume that can be difficult to manage.
“The great part of the innovation here is, with the digital solution, we’re 100 percent scalable.”
For more “complex” transaction disputes, CBA is using AI “to serve up to the customer a much richer, much more detailed and higher quality conversation”.
The AI can fill in information gaps not provided by the customer to make it easier to lodge complaints.
Messaging gets AI treatment
Messaging was initially launched in the CommBank app in 2018.
Sullivan said the bank is now launching “a generative AI-powered messaging service direct to our customers.”
Customers are given the opportunity to ask if they would like a generative AI-powered response if they want more insights beyond predetermined answers.
“If not, then we still have the option to bring that customer direct to an agent.”
Banker workbench
CBA has built a “banker workbench” to help orchestration of customer interactions using AI and data, enabling bankers to more timely and relevant conversations with customers.
The idea is to enable bankers to get on the phone and engage customers with proactive and pre-emptive insights.
Rebuild and modernise
CBA executive general manager of retail technology Vicky Ledda said the bank is continuing to “rebuild and modernise its technology estate for AI”.
She said that staff are being upskilled to lead, build and work with AI.
“We’ve recently appointed our first distinguished AI scientists who, together with our distinguished engineers, are solving the most complex technology problems at CBA through building with AI,” she said.
“We’re upskilling all of our engineers, empowering them with skills, tools, and infrastructure they need in order to build gen AI-enabled solutions.”
The bank has also created an ethical generative AI toolkit that includes 11 guardrails.
“These are very specific to generative AI risks. One of the guardrails, for example, mitigates against the risk of jailbreaking, which is a common practice used to try and manipulate large language models in harmful ways,” Ledda said.
Ledda said also touched on the environmental impact of AI “which is why CBA has purchased the equivalent of 100 percent renewable energy for our Australian operations.”
“This also includes CBA AI-related energy [use] in our own data centres”.