Bendigo Bank has a pipeline of more than 3000 ideas to use agentic AI to “make the organisation better” and is running them through a series of gates to determine prioritisation.
Bendigo’s chief technology officer Kieran O’Meara.
Chief technology officer Kieran O’Meara told the Google Cloud Summit Sydney that the 3000 ideas were crowdsourced from its workforce of “over 5000” staff in the space of nine months.
He said that this number of ideas was only possible because the bank put AI tools – notably Google Cloud’s Gemini – into the hands of all staff and invested in upskilling and education.
“What we didn’t do was we didn’t put [AI] in a lab. We didn’t run experiments in the traditional sense,” O’Meara said.
“We very deliberately went wide and we started at the top with education, including the board and the executives, because these leaders needed to know how to lead this work.”
O’Meara said the ideas were not for enhanced “personal productivity or team level innovation”, although AI had a role to play there.
Instead, he said, the efforts needed to have an impact for Bendigo Bank as an organisation.
Now, the focus has shifted on what to fund and progress.
O’Meara said that the ideas were being progressed and considered for funding only if they passed a series of gates.
These gates include whether the idea really requires agentic technology or not.
“The single most common thing that happens is we have hammers looking for nails,” he said.
“Every idea is an agentic idea. Actually, the reality is about 80 percent of what gets pitched is not really an agentic problem. It’s not really an agentic opportunity. It’s a deterministic opportunity. We can solve it in other ways.
“But the things that we choose to go forward with are most commonly both: they require both intelligent reasoning or judgment and deterministic work to achieve the outcome.”
Other checks include whether an agentic implementation of the idea would be auditable and, if something went wrong, what the repercussions might be; and the economics of doing the activity agentically.
“I have seen real examples where the agentic workflow is more expensive than doing it the way we do it today,” O’Meara said.
“In one particular call centre example, the cost of the interaction agentically was going to be about $9 versus the human-supported interaction [cost] of about $3.50.”
Another gate is whether the organisation is actually ready to progress the idea.
“We’ve fallen into this trap a few times. Do we have the data? Can we build the integrations? Do we understand the semantics? And do we have that multidisciplinary team to achieve the outcome?” he said.
“[This is] essentially how we test what comes in and how we decide what to move forward with to put the real money into achieving the outcome.”
Ry Crozier attended Google Cloud Summit Sydney as a guest of Google Cloud.
