- Why giving staff access to an AI tool without cleaning your data first is an expensive mistake
- The AI discussion remains frustratingly surface-level–both at the enterprise and national level
There is a story Cassandra Goh tells that says more about the state of AI adoption in banking than most conference keynotes manage in an hour. A large organisation, she recounts, had deployed an internal AI chatbot, trained on its own document repository. On launch day, someone asked a simple question: who is the CEO of this company? The chatbot had no answer.
It was not a failure of the technology. It was a failure of preparation, of the unglamorous, painstaking work of tagging documents, mapping metadata, and giving a system the context it needs to function. Nobody had done that part. They had gone straight to the pilot.
“That is not what you call an AI POC project,” Goh says, without a trace of condescension. “It is a basic cleanup that you have to do for however many years. Either that or junk up.” Goh became Group CEO of Silverlake Axis on July 1, 2024, stepping up from her role as Deputy Executive Chairman. She was supposed to take over in Jan 2025 but the elevation was brought forward with the board citing its confidence in her leadership.
She inherited a company that sits at the nerve centre of financial services infrastructure across Southeast Asia, one that powers the region’s banking quietly and without fanfare, the kind of company that does not make headlines precisely because nothing goes wrong.
Founded in 1989 by her father, Goh Peng Ooi, a math nerd who had walked away from IBM Malaysia to launch Silverlake. It was a gutsy move by Goh, believing he could compete on a hardware and software basis building a world class tech company to compete against global tech companies to serve banks in Malaysia. It certainly paid off as today Silverlake serves not banks in Malaysia but 40% of the top 20 banks in Southeast Asia. Its core banking systems process transactions for institutions across 80 countries.
It does not make headlines, working in the trenches of banking infrastructure to ensure systems and networks work. But if it beats Chinese, Indian, American tech leaders to win the large Maybank Group contract to replace its core banking system which represents part of its RM10 billion budgeted, five-year rebuild of its technology foundation, then Cassandra will be in the spotlight, and deservedly so.
For the Gen-Y CEO (she is in her mid 40s) leading the company in a new era, the question facing her is, what does it mean for the company, and for the banks it serves, to make AI work. Today, they may be freaking out over Anthropic’s Mythos model, which has ushered in the era of Zero-Second Threat, or over Google announcing last month that it found a group of criminal hackers had used AI to find an unknown bug. Tomorrow, who knows what threat will emerge to keep the banking c-suite up at night.
Not broken enough to fix – until now
When Goh joined Silverlake as Deputy Executive Chairman in October 2023, generative AI had just arrived in the public consciousness. ChatGPT had been released weeks earlier, and the financial industry was doing what it always does with new technology: it was watching, debating, and quietly panicking.
“There was everything between: if we don’t adopt AI, we will die, to, I use AI to book my travel and it is still so stupid, so are you really telling me it is going to change my back office?” She laughs, then turns serious. “I am not the kind of person who will tell my customer that technology will save you. It is just an enabler. If you do not know how to be creative in your business, technology is not going to magically make you creative.”
The real problem, as Goh sees it, is not AI readiness in the narrow sense. It is something that has been accumulating for decades. “If you are looking at banking systems within this region, they are all between 20 to 30 years old. They were never designed for this.”
A core banking system installed in 1997 was a remarkable achievement for its time. Nobody sat down in 1997 and imagined that by 2025, an AI would be expected to interpret unstructured documents, navigate complex workflows, and make decisions with human-equivalent context – all without a screen to look at.
This is why Goh argues that most of the discussion around AI adoption misses the point. The real work is not choosing an AI model or running a proof of concept. It is the plumbing: re-architecting applications, cleaning data, establishing metadata standards, and building the kind of structural context that allows machine-to-machine communication to function without a human in the loop.
“You develop applications today for you as a human. You see the screen flow from A to B to C. You touch the screen. You intuitively know you have to press this button. So it is implied that there is a human moving things, and the human has context in their mind. How do you bring that context into the machine so that when you don’t see the screen, it still works?”
This is the harder conversation, and it is one Goh believes the industry is still avoiding. The AI discussion, she says, remains frustratingly surface-level–both at the enterprise and at the national level. “It’s a very surface-level discussion. AI is here. It’s probably going to do something. Let’s go do something.”
The FOMO is real even in banking
Goh was not always convinced banks would be susceptible to AI hype. They are, after all, institutions run on risk management. They answer to regulators and shareholders who demand dividends, not R&D investments. But she has changed her view.
“Everyone falls into FOMO. Banks are run by humans.” She points to the digital banking wave years earlier, when incumbent institutions convinced themselves that neobanks would make them irrelevant. “Three or four years ago, everybody was saying: digital banks will kill us, we will be irrelevant. And then you ask them today what they think about digital banks? And they are not sure.”
With AI, she sees a similar cycle playing out, though with higher stakes and deeper structural implications. The difference is that this time, there is a legitimate foundation underneath the hype.
AI adoption in Malaysia’s financial sector has accelerated markedly. Bank Negara Malaysia’s data from a Aug 2025 published report shows that 71% of banking institutions operated at least one AI application by the end of 2024, up from 56% a year earlier. More than 60% of banking and insurance institutions in the country view AI as a strategic priority over the next one to three years.
But adoption and readiness are different things. The same research points to data quality as the central obstacle – AI models, as PwC Malaysia has noted, are only as effective as the data on which they are built. And for institutions running legacy systems built before digital records were standard, that data layer is fragile.
Goh sees this challenge across her client base. The biggest banks have the capacity to experiment, they can run multiple proof-of-concepts across departments simultaneously and absorb the cost of failure. Smaller institutions face a harder calculation. “If you are unclear on the ROI, how much are you going to burn doing experiments?”
She also pushes back on the belief among some tech execs that AI costs nothing to start. “There is an underlying hangover from cloud and software-as-a-service,” she cautions. Infrastructure that looks cheap in the early stages accumulates into a substantial recurring cost once you reach scale in your AI adoption. Compute, licenses, data governance tooling – it all compounds.
An on-premises machine is a fixed cost. Cloud-based AI infrastructure is a variable one, and not always in the direction that budget projections assumed she cautions.
Singapore runs. Malaysia watches. Indonesia moves differently.
When asked about how the landscape looks across Malaysia, Singapore and the Indonesian banking landscape, Goh’s answer was measured. Singapore is the fastest mover. “They are more willing to try and go.” The combination of competitive pressure, access to talent, and the city-state’s positioning as a global financial hub creates conditions that push banks there toward faster adoption cycles. It is, as Goh acknowledges, a structural reality.
Singapore banks are not competing just against regional peers; they are competing against institutions in London, New York, and Hong Kong.
Malaysia, she describes as being an observing market. “Malaysia has always been the kind that likes to see who makes the first move.” Asked whether this poses a long-term competitive risk, she is balanced. Running too far ahead carries its own costs, both financial and operational. But waiting too long means losing ground that cannot easily be recovered.
“You have to know yourself. How fast you can react. And whether you have the ability to see when the moment comes.”
For Indonesia, she declines to make generalisations. The country’s size and complexity, its geography, its diversity of banking markets from large state-owned institutions to community rural banks, makes a single characterisation difficult.
What she does say is that the focus there, as in Malaysia, is fundamentally on tech refresh and digitalisation. AI is part of the conversation, but it is not leading it. What matters more in all three markets, she argues, is not the pace of AI adoption per se but whether the foundational work is being done. System upgrades. Data architecture. Application rationalisation. These are not AI projects. They are the prerequisite for AI projects.
The board has a question, CTOs have their answer
In the boardrooms of her banking clients, Goh says a consistent question is emerging: do we have a plan for AI? Boards are not driving specific programmes. They are asking executives to put a roadmap on the table. “The sense I get is that we are all waiting for the first tangible step to take,” she says. “It is new. What is a tangible AI playbook?”
The industry is awash with frameworks and benchmarks, but translating them into executable roadmaps for institutions with decades of technical debt, heterogeneous infrastructure, and significant regulatory constraints is a different challenge entirely.
CTOs, she says, are not being handed larger budgets on the back of AI enthusiasm, not in financial services, at least. The investor dynamic in banking does not reward R&D the way it does in technology companies. Shareholders want dividends. The pressure to invest in AI capabilities must be justified against near-term returns, and that calculus is difficult when ROI timelines for infrastructure transformation are measured in years.
What AI has done, however, is something almost paradoxically useful: it has given CTOs a new argument for investments they have needed to make for a long time. Modernising a core banking system that was not broken, technically, was a hard sell. Now, with AI requiring clean data architectures, modern APIs, and structured service layers, the argument writes itself.
“If you are a CTO coming out for a budget, and someone asks why you want to do this – if it’s not broken why fix it – at least now you can say: it’s broken because AI is here. I think it allows people to start being willing to talk.”
Silverlake’s work has never been more relevant
As MDEC projects a US$115 billion AI opportunity for the Malaysian economy by 2030, what role does the financial sector play?
She does not pretend she knows the answer and highlights that banking systems are aging and that it is impossible to deploy AI on infrastructure that was never designed to expose its data in a structured way.
“You can only prepare for what you know. There is definitely a devil in the unknown when it comes to AI, because I cannot tell you what the right bet is six months down the road, with the models improving so rapidly.”
The implication is that certainty is the wrong goal. The right goal is building institutions and systems that can adapt when the unknown arrives. “You have to at least know what you can do to be prepared for most of it.”
For Silverlake, this is where the business case sits: not in selling an AI product, but in helping banks build the structural conditions under which AI can be meaningfully deployed. System architecture. Application renewal. Structured services. The company’s positioning under Goh is that this is what they have always done and that it has never been more relevant.
The bubble question
Are we in an AI bubble? Goh does not dismiss the question. She acknowledges that some of the current cohort of AI startups will not survive the next consolidation wave. That as foundation models become more powerful, the differentiation available to companies built on top of them will narrow.
“The smarter the LLMs get, the more you can just copy and paste and build your own.” But she draws a distinction between the bubble in AI companies and the permanence of the shift AI represents. She notes that the dot-com bust of 2001 did not undo the internet. It eliminated companies that had not built real value. What survived, e-commerce infrastructure, digital payments, online communication, went on to transform entire industries.
“When people say the AI bubble, there were things that survived the dot-com bubble,” she says. “I think the change that comes with having an LLM accessible to people is here to stay. Your expectations change. What you need to learn as an employee changes. How much knowledge you have on your palm changes.”
Building for sustainability in an AI defined era
Inside Silverlake itself, Goh has been running her own version of the transformation she prescribes to clients. She describes a significant reorganisation that she initiated when she took the helm, not a cosmetic restructuring of names and org charts, but a substantive push to change how the company thinks about and builds its products.
“I came on to do this,” she says simply, when asked about the timing. The impetus was the same recognition that confronts every enterprise software company in the AI era: that applications built for human interaction, designed around screens, flows, and implied human context, need to be redesigned for a world where machine agents will increasingly be the primary users.
She is unsentimental about the discomfort this causes. “Change is hard because then everyone is like, what is the next step.” She describes telling her developers that they need to build applications in new ways – ways that make their current work harder but that will pay off in a future where AI-driven orchestration is the norm. “They will say: Why? And I say: We want to go like this. It will make your life a little bit harder. But it will pay off.”
The question of whether this transformation has affected Silverlake’s headcount gets an honest answer: “Things go up, things go down.” She does not point to AI-driven automation as a driver of job elimination at Silverlake.
What keeps her up at night, she says, is the question of whether the direction is right. Not the execution, she trusts her team on that. But the directional bet. “Directionally correct. But each step, be ready to pivot.”
AI was used to generate the first draft before the writer and an editor worked on the published version.
Related Articles