Bendigo and Adelaide Bank has used generative AI and MongoDB to rewrite the code for a retail banking application, opening up a new pathway to modernise and move some of its older applications into the cloud.
.jpeg&h=420&w=748&c=0&s=0)
The modernisation of its Agent Delivery System (ADS) - teller software used by pharmacies and other non-bank businesses (“agents”) to offer banking services in their communities - is being promoted as a global-first, performed in collaboration with MongoDB.
Like many Australian financial institutions, Bendigo and Adelaide Bank is in the midst of a cloud migration but has found not all applications and workloads can be lifted-and-shifted or modernised easily.
“Older applications can really stifle a cloud migration,” the bank’s chief information officer Andrew Cresp told iTnews.
“I’ve had experience at two organisations now where you get to between 40 and 60 percent of your workloads [migrated], and you look at the remaining applications and [conclude], ‘this is going to cost us a fortune. How are we going to create a business case that works?’
“ADS was really going to be one of the examples of this.”
The solution came in the form of a combination of MongoDB Atlas and generative AI tooling created by MongoDB”s professional services organisation.
“The MongoDB team were primarily interested in helping us move the [underlying] database [for ADB] onto their capability, which is ideal for cloud. We actually got through that quite quickly, in … three weeks,” Cresp said.
“This is where the exercise really changed. We had everyone there, they said moving the database is interesting, but moving the application is the hard stuff. Why don’t we have a crack at that?
“So, the joint team worked together on how to refactor that application based on the way it was doing database calls. That’s why this global-first use case is a really powerful one because we actually needed the MongoDB [team] to think in the MongoDB way.
“They’re thinking about how the application ‘talks’ [to the database and other systems] and then working that back through, which I think is the critical differentiator from other AI programs.”
For Bendigo and Adelaide Bank, a key outcome as well is confidence that generative AI can do a lot of the heavy-lifting around application modernisation.
“This wasn’t a ‘copilot’, just helping our developers be 30 percent more efficient,” Cresp said.
“It actually rewrote the app, created documentation … and automated testing capabilities for the app.
“This app modernisation use case I think is a game-changer for industry, generally.”
While acknowledging some initial scepticism and “healthy cynicism” around asking generative AI to rewrite an application, Cresp said the output was 90 percent of the way there.
“We were quite surprised at the quality of the 90 percent,” he said. “It [just] interprets some things incorrectly, which you obviously still need humans [to correct].”
Given the banking use case and regulatory restrictions, the generative AI was private to Bendigo and Adelaide Bank.
“The large language model we were using here is internal,” Cresp said.
“We were saying to it, ‘go learn how to write good APIs from our APIs’. We didn’t need to send anything outside our boundaries.”
The entire modernisation was run over a period of three months.
MongoDB confirmed that the typical engagement model is over 14 weeks, comprising seven fortnight-long sprints, with the intention to get the modernised application to pre-prod in that timeframe.
Cresp said there was “a big discussion [internally] over which application” to run through the novel modernisation process first. “We didn’t want to bite off something too large,” he said.
Still, a business-critical core application was selected - and the end result has Bendigo and Adelaide Bank confident that it can now use the same method to modernise some bigger targets.
“We are doing our branch teller system next,” Cresp said.
“We are currently working on our entire branch teller system [and] our payments platforms, [and] we’re using this [modernisation] approach right now.”
The ADS modernisation had also generated considerable interest internally as to how much it can be replicated across the application estate.
“Certainly now, our biggest problem is everyone wants to use this capability for those applications and we’re just trying to manage the excitement,” Cresp said.
“We’ve got a laundry list of the next applications we want to use this process on and the team that were involved in this are really enthusiastic.”
Cresp said that the bank had also looked at other unnamed generative AI tools as part of the ADS modernisation, before advancing with MongoDB.
Aside from enabling modernisation to happen quickly, the bank said the method was also cost-effective compared to other more intensive alternate approaches.
The intent is to make more use of this application modernisation method, while also keeping one eye on the rapidly evolving generative AI space.
“The challenge with generative AI is they’re just leapfrogging each other the whole time, so I think the trick is not getting too hung up to say this is our answer for everything because in three months' time it might change,” Cresp said.
