Westpac is hoping at least 1000 engineers will be coding with assistance from generative AI by the end of the year, signalling the growing importance of the technology to the bank.
Chief technology officer David Walker said a recent in-house AI pair programming experiment resulted in a 46 percent productivity gain, with no reduction in code quality, from software engineers that were aided by generative AI, when compared to the results of a control group that performed the same tasks exclusively by hand.
Walker said he was surprised by the “positive result”, not expecting productivity gains to be so high, or for teams to embrace the technology, given “healthy scepticism” going into the experiment.
Walker’s goal with the experiment was to test if generative AI pairing could improve the working experience, productivity levels and quality of code produced, compared to more standard coding works.
“They were the three things that we wanted to prove as part of this experiment,” he told the iTnews Podcast.
“Without doubt, all three were proven successfully. We’re very excited about the results and very pleased that we could prove our [hypotheses].
“Not all experiments prove out as successful as this one.”
To shape the experiment, Walker gathered 60 developers from across the bank and divided them into four groups, with each assigned “a nice, interesting task”.
One group was to use standard coding techniques while the other three were able to leverage generative AI tools from Microsoft, Amazon and OpenAI.
Walker added the trial produced reusable code and also demonstrated how much value “boilerplate” code from a generative AI tool could add.
The experiment was also time-limited to 24 hours: a ploy to understand just how easily AI pair-programming tools could be harnessed to produce useful code, although Walker noted this wasn’t one of the formal outcomes he was seeking from the pilot.
“These tools on the surface appear to be so easy to use. You ask questions and up pops code,” Walker said.
“It seemed fairly straightforward, so we wanted to see what sort of learning curve there was on it.
“It turned out to be a lesson in terms of how easy it is.”
Ramp up plan “underway now”
Experimentation with generative AI also enabled the bank to put “hearsay information” about the technology into a Westpac-specific context, which enabled the bank to “then move forward with confidence.”
“We’ve seen many forms of [AI], and the benchmark of what someone calls AI or artificial intelligence is constantly changing,” Walker said. “Generative AI was a reasonably big step forward.”
Walker said Westpac has been using AI in some form since “the late ’60s, early ’70s”, coinciding with the advent of the automatic teller machine (ATM).
“That’s probably the first, type of AI that we saw,” he said.
“We’ve been adopting AI for many years and generative AI had been the latest iteration [of that journey]”
A ramp-up plan around generative AI is now underway, Walker said.
He predicts 1000 engineers will be coding with the assistance of generative AI by the end of the calendar year.
Walker said that AI pair-programming is “now actually being used in real-life coding rather than just as an experiment”.
The work is benefiting its main mobile banking app and consumer-facing services and features.
“Part of that really is now about ensuring that we’re streamlining the use of it, making sure the engineering experience is great [and] working with the teams to adopt it.”
He added another use case is assistance in writing unit tests.
Talent focus
More broadly, Walker called out Westpac’s long-term focus on technology. “It’s about making sure that we keep on the forefront,” he said.
Walker characterised the benefits of that as being around customer service and aiding the bank in the ongoing war for talent, by “making sure we have all the latest tools and environments to work on.”
“If you are an engineer out in the market, then clearly you want to go to companies that are forward-thinking,” he said.
“That’s a big part of doing these sorts of generative AI experiments, to test-and-learn and then adopt them where it makes sense.”
Engineers that participated in the generative AI experiment were surveyed for their feedback and sentiment.
“By the end, I think there was a clear consensus that this was something they wanted to use and get into,” he said.
“In fact, the 60 engineers that are using it today are the privileged ones, and every day my team’s getting requests from other engineers around the Group saying, ‘When can we get in? When can we get into it?’
“There’s definitely a positive vibe around it.”
From the survey, 83 percent of engineers saw generative AI as a learning and assistance tool.
New ambitions
The testing has unleashed a multitude of use cases for generative AI that are now underway, with many people having “woken up” to other possibilities across the bank.
Westpac is now applying generative AI to its internal data, “particularly around things like HR policies and processes”.
Customers can expect the Westpac mortgage origination process experience to shift with exploration around “how does AI support them in this process to make sure that the information that they’re providing is full and complete, so we’re not wasting their time.”
The aim is to cut down mortgage processing time and improve straight-through processing.
“Financial markets are another area we’re looking at,” Walker said.
“Bankers have a lot of products and a lot of complex policies. We want to make sure all this information is at their fingertips when they’re talking to customers.
“Having a generative AI model learn all that information and then again, let the bankers just go and ask it questions and then get the answers back is a really powerful thing.”
Another area of interest is in Westpac’s Microsoft 365 environment.
“We’re part of an early-stage program working with Microsoft on how we can use generative AI inside Microsoft Word, PowerPoint and Excel, as again, a bit of an assistant sitting in there to support [users],” Walker said.
He added the bank is also looking into applications across fraud and cyberspace as well.
Walker also noted that the bank is seeking to balance use cases with ethical and responsible use of AI.
“We’re very conscious these things need to be right. We have a very strong view on that, and part of everything we do is making sure that we do no harm,” he said.
While experiments are growing, the bank is ultimately taking a “very cautious” approach via its use of pilot projects to better learn use cases are safe for staff and customers.