Track · AI in Product5 min read
CBA’s $90m AI bet puts context, control and focus ahead of scale
CBA’s CAIO says AI strategy must start with context, not models: four priorities, five enablers, $90m in upskilling and controls for every agent in production.

Context
$90 million is CBA’s investment in upskilling teams through its skills-and-capabilities enabler.
CBA’s four AI priorities are customer experience, fraud and scams, faster delivery and reimagining work.
Its five enablers are one AI platform, an AI control plane, token optimization, context and knowledge, and skills and capabilities.
CBA registers every AI asset and requires pre-deployment testing, real-time drift monitoring, after-the-fact outcome checks and independent validation.
Commonwealth Bank (CBA) chief AI officer Ranil Boteju says the bank is investing $90 million to upskill teams as part of skills and capabilities, one of five enablers in its AI operating model. Onstage at Momentum Media’s FST Future of Financial Services, Data & AI conference, he laid out the wider model: four business priorities, shared capabilities and a registry for every AI asset.
Who owns AI when the CEO is the real chief AI officer?
The chief AI officer (CAIO) title is still finding its place. CFOs, CMOs, CIOs and CISOs all have stakes in AI, and their established responsibilities are clear. CAIOs face a less settled mandate: they may be influential and respected while holding different degrees of authority.
Boteju runs a centre of excellence and builds shared capabilities, but he does not run a vertical silo. He drew a parallel with chief digital officers, whose separate digital teams left everyone else outside digital work. His conclusion: AI transformation cannot be constrained to one team.
Each business unit must lead its agenda, supported by common capabilities built by the CAIO. The CAIO must also manage up to the CEO, collaborate across the C-suite and give teams top-down inspiration for embracing and extending the vision. Boteju said: “I am the chief AI officer, but I’m one of many.” In his view, the CEO is the real chief AI officer; if CEOs do not understand AI’s importance, the organization has bigger problems.
Takeaway: The title supplies coordination and standards, while the CEO sets direction and business units own adoption.
Can context become the moat?
Boteju’s first lesson addresses a harder competitive problem: when banks use the same AI technologies or large language models, access to the model no longer differentiates them. “In a world where all of the banks are using the same models, what will differentiate us is the context, the organizational memory, all of the tacit knowledge that sits in people’s heads,” he said.
Context means patterns, experience and hard-earned wisdom that employees may never have written down. Standard operating procedures often remain high-level, with much of the real knowledge held in people’s heads.
Some of it is scattered through unstructured data, but Boteju argued that analysis, not a lack of data, is usually the gap. Asked when data would become good enough, he said most financial-services organizations have no excuse to wait.
For product managers, the practical test is blunt: how much does the organization know in writing, and how much could it hand to an agent tomorrow?
Takeaway: Before buying another model, make unwritten knowledge visible and usable by agents.
Why does CBA keep its AI strategy to one page?
Boteju presented the thinking on simple slides. He says the reason for doing AI must be clear rather than a corporate bandwagon. CBA has chosen four priorities:
- Customer experience
- Fraud and scams
- Faster delivery
- Reimagining work
The next question is how AI can deliver against each priority at scale. CBA groups the answer into five enablers, shown on one slide:
- One AI platform
- An AI control plane for centralized oversight
- Token optimization, meaning control over model usage and spend
- Context and knowledge
- Skills and capabilities, backed by the bank’s $90 million upskilling investment
Boteju described an earlier enterprise pattern as 1,000 flowers blooming: duplicate projects and proofs of concept that failed to scale. His prescription is a narrow set of use cases, ideally expressible on one page.
Takeaway: A strategy becomes operational when every use case supports a stated priority and draws on shared capabilities.
How do you control agents you cannot fully predict?
Boteju pushed back on the dominant AI-industry claim, often voiced by large language model leaders, that nobody can control AI or the agents its creators build. CBA instead uses a layered machine defense. The bank registers every asset: at any moment, Boteju can identify every AI agent underway, live in production or being built.
Nothing goes live without pre-deployment testing. Teams monitor drift—unwanted changes in system behavior—in real time, check outcomes after the fact, and have an independent team validate the work.
Control also shapes the five enablers. A single gateway manages token spend, with caps for engineers and smaller, cheaper models for less intensive tasks. The platform and control plane provide oversight, while Boteju identified context and knowledge as the one thing the bank most needs to protect.
The operating principle is to know what can and cannot be controlled, then plan for both.
Takeaway: Visibility, testing, monitoring and independent validation keep AI agents inside a managed operating model.
What should product leaders do on Monday?
CBA’s model offers a practical test:
- Make CEO sponsorship explicit and name a business-unit owner for each material use case.
- Put priorities, use cases and shared enablers on one page.
- Register every AI asset, including agents in development and production.
- Capture context and organizational memory in forms agents can use.
- Set pre-deployment tests, real-time drift monitoring, after-the-fact outcome checks and independent validation.
- Route model use through one gateway, cap engineer token spend and use smaller models for lighter tasks.
The title alone will not produce transformation; success depends on the whole team understanding and communicating the required change. Leaders should also challenge the industry’s fear, uncertainty and doubt narrative rather than accept it as the operating plan.
As AI moves deeper into production, product practice will increasingly turn on narrow mandates, usable organizational context and controls that teams can audit.
via cio.com (Source)
More from Daniel Okafor
Show full bio
Staff writer covering technology and product news in product management at Roadmap File.
60 articles