CommBank iQ is putting AI to work on its core dataset

CommBank iQ is bringing AI to bear on its main banking data asset, after security, governance and risk concerns meant the technology was used almost everywhere except in the core.

Using AI to interact with its core banking behavioural dataset is “fundamentally changing” analytics workflows behind-the-scenes while getting answers to clients faster, former chief executive Makenna Ralston told the Snowflake Summit 26 in San Francisco last month.

A joint venture with Quantium, CommBank iQ commercialises a de-identified transaction dataset from CBA, using it as the basis for insights into consumer spending behaviour.

Ralston - who is staying with Quantium but is London-bound - said that CommBank iQ hit a roadblock in its AI ambitions over the past year.

A year ago, Ralston was “convinced” that going “all-in” on AI would unlock capacity in the team and enable CommBank iQ to “put more of our insights into the hands of more diverse leaders to make better decisions.”

But with a business model linked to CBA’s “crown jewels” - its data - Ralston said that security, governance and risk considerations ultimately prevented the organisation from using available AI tools on its core data asset.

“We uphold a bank's security posture. That makes sense; we are touching and talking about banking customer data … the crown jewels of that organisation,” she said.

“We [also] operate in a joint venture governance structure. Risk and compliance and governance controls on the best of days are complex.

“And for all of those reasons, it meant that everything we were excited about, all the AI tooling that was coming about, it was all awesome, but none of it could actually touch our data.

“So, what was happening is that our team was getting smarter working around the data. They were using AI every single day for everything else that they could do that didn't have to touch the data: AI for strategy work, AI for customer communications, AI for all of the work that goes into knowledge work but isn't on the core asset.

“The problem was our AI tooling and our AI excitement … could not reach the fundamentals of [the] business. 

“The promise of AI for us just had not materialised in the way that actually reshaped the core of what we do.”

This limitation, according to Ralston, was being felt by the broader CommBank iQ business, in its responsiveness to client briefs.

“Our analysts were still following the same long workflows every time the client brief came in, [and] our client teams were still not able to answer some of those insight questions in the room with the client because they needed to go back and raise a query,” Ralston said.

“All of these talented people were getting pulled into mundane, routine, analytical tasks despite all of the AI investment and innovation that we were making. 

“AI hadn't changed any of that yet. It was only happening around the edges.”

The company was also at a limit capacity-wise and had to turn down work.

“New work came in. Exciting work, good client, good problem to solve. We really wanted to get involved in this. The type of pipeline moment that certainly as a CEO, you usually celebrate - and we couldn't meet it. The team was simply at capacity,” Ralston said.

“Our team is brilliant, but … very finite, and when that exciting work came in the door, it hit a fully at capacity team. They had zero headroom, and so as CEO, I had to turn the work away.”
What’s changed for CommBank iQ is better AI tooling that is native to the ecosystem it uses for its core data analytics - Snowflake.

“Something genuinely changed - not in our strategy, in Snowflake's roadmap. The new generation of Snowflake capability was the first time for us that we got material AI horsepower right in the core, right where it mattered for us,” Ralston said.

“So, Snowflake's roadmap met our reality right when we needed it to, which means that we now have a genuine shot at rewiring our business so that we can scale.

Specifically, Ralston raised two Snowflake AI tools that were recently rebranded - CoWork, formerly known as Snowflake Intelligence, and CoCo, formerly Cortex Code.

CoWork is an agent that can query the dataset, whereas CoCo is more used in the backend for building data pipelines and machine learning models.

Ralston said CommBank iQ was given early access to the two AI tools.

“Specifically, [Snowflake] allowed us to test early and come on that roadmap journey with them,” she said.

Tourism insights for NSW

While behind-the-scenes efficiencies have been realised, Ralston was most excited by what AI would mean for the intelligence firm’s clients, and for staff in client-facing roles.

AI, she said, is “reshaping how we work completely”.

“Our clients don't really care about analytics. Analytics is just a translation layer between data and answer, and historically, that translation layer has taken too long,” she said.

“Think about it. The client asks a question, the client lead writes it up, then goes to an analyst. The analyst writes the code, then they run the query, then they bring the answer back, then the client lead translates that into an insight, story or presentation and delivers it back to the customer. 

“All that means is that the only questions that get asked are the ones that are worth waiting for. So the smaller questions, the curious questions, the half-formed thoughts that might actually turn out to matter more than whatever the strategy deck says you should ask, those don't get answered because they don't get asked.”

Ralston said that with CoWork in place, non-technical users can query the organisation’s foundational data “directly in natural language in the room with clients”. 

“We're using Australia's largest state, NSW, as a testing ground for Snowflake CoWork on our visitor economy insights,” Ralston said.

“The team had only had the tool for maybe a month, and in that time, a real government client asked, "What is going on with this unusual spike here that I can see in the transaction growth? What's all this about?" 

“They knew it had been school holidays, so usually you see some more tourism there, but they weren't quite sure why that was happening differently this year, what was driving that. “Honestly, it was a pretty minor question, and in the old world, that client lead would have taken that away, given it to an analyst, and come back after they'd worked it up, maybe a week later, maybe after the client stopped caring. 

“However, the non-technical client lead with Snowflake answered that question directly in the room.”

Ralston said the growth was attributed to school holidays and a new tourism business that had opened.

That insight, she said, afforded the government client “an opportunity to go and speak to that business, to think about funding for that business, to bring them into the catalogue of experiences that they're offering as the tourism leaders in that space. “

“That's a real actionable insight. A small question, a curious question that got answered in the room, not a week later,” she said.

“If you multiply that out by every question that every client asks every day that previously went into a queue, you're starting to look at what that democratisation of access to our data really looks like in action.”

Whilst early days, Ralston said it is “a strong preview of what's to come.”

“Our job now is to take that and apply the business logic to all of our other different industries so that we make sure this capability is right across our business,” she said.

“We're working on that now.”

Virtual agents

More ambitiously, CommBank iQ is building “virtual analysts” on CoCo, which Ralston described as “a digital team member that handles route analytical workflows end-to-end.”

“Give it a scope, give it a client problem, and it'll do the thing right through to delivery,” she said.

“This is not a tool that my analysts use. This is their new teammate. 

“We are now moving to a hybrid human and AI model in our business, [with] humans and AI doing exactly what each of them should be doing.

“The analysts will set the brief, they will make the judgment calls, and they will hold the client relationships. The virtual analyst will do the work that the analysts were doing on nights and weekends.”

Virtual analysts could be in production as soon as this month, Ralston said, initially in one team before scaling out from there.

“It's not fully there yet,” she said.

“We're still testing the edges, but we are weeks, not months, away from reshaping how we work completely with this Snowflake capability. 

“This was the dream we had last year. We just couldn't get our hands on it.”