How Bank of Singapore Is Rewiring Compliance Around AI and Risk Culture

Bank of Singapore’s Head of Platform & Analytics Kelvin Chiang told a private-banking audience last week that the bigger story of Bank of Singapore AI compliance work is not faster reports. It is a rewire of how the institution understands risk across the entire client lifecycle, from the first prospect meeting to the moment a customer’s circumstances change.

Speaking at the Hubbis Digital Dialogue webinar “Holistic Financial Crime Compliance in the Age of AI” on 1 July 2026, hosted in partnership with IMTF, Chiang framed the question as cultural, not technical. “Artificial intelligence (AI) represents something larger than compliance efficiency – it’s about redefining what’s possible in banking,” he had earlier told Asian Private Banker. The rest of the argument traces what that “larger” looks like in practice: a source-of-wealth tool already shipping, a pre-onboarding risk filter under design, a cross-divisional intelligence layer still being built, and a claim that AI’s deepest value is shaping how relationship managers think.

The Client-Lifecycle Frame That Anchors the Whole Argument

Chiang opened the session with a working definition of compliance as a lifecycle discipline, not a checklist. The bank has to evaluate risk before onboarding, monitor it once a customer is live, and keep re-evaluating as circumstances change. The IMTF webinar listing set the same scope: “Building a 360° view of customer risk across the compliance lifecycle” and “breaking down silos between KYC, AML, sanctions, fraud, and investigations” are the explicit topics of discussion.

That lifecycle lens matters because risk is not static. A client can pass onboarding and turn higher-risk months later because of transaction behaviour, adverse media, a change in beneficial ownership, or a geopolitical event touching their industry. KYC risk scoring, country exposure, sanctions posture, transaction history and adverse news all feed the same picture, and Chiang’s argument is that AI has to read them together rather than in serial handoffs between teams.

The framing echoes how the Monetary Authority of Singapore (MAS) now talks about the work. The regulator’s consultation paper on AI Risk Management, published last year, treats AI oversight as a lifecycle problem in its own right, from model inventory and validation through ongoing monitoring and decommissioning. Chiang’s argument is that the same lifecycle discipline that MAS expects of AI models is what AI is supposed to deliver for clients.

Source of Wealth as the First Concrete Win

The most concrete proof point is the Source of Wealth Assistant, or SOWA, that Bank of Singapore rolled out on 10 October 2025. According to the bank’s media release, the agentic AI tool compresses source-of-wealth report drafting from the usual 10 days to as little as one hour, while validating client information against OCBC and Bank of Singapore databases using benchmarks such as salary and company revenue.

That compression is not the win Chiang is selling. The win is consistency. Relationship managers still have oversight, reviewing and refining every AI-generated draft before it moves to internal review teams, but the first pass now looks the same regardless of which analyst or which office produced it. In a private-banking world where source-of-wealth corroboration was named explicitly by MAS as a deficient control in the August 2023 laundering case, that consistency is the regulatory point.

Agentic AI pushes the envelope further by enhancing efficiency, accuracy and consistency in decision-making. With AI integrated into the source-of-wealth reporting process, relationship managers can shift their focus from manual documentation to meaningful client engagement and risk assessment.

Kam Chin Wong, Global Head of Financial Crime Compliance at Bank of Singapore, made that case in the bank’s announcement. The tool sits on Bank of Singapore’s private cloud, runs inside the OCBC Group Model Risk Governance framework, and is monitored by a centralised guard-rail system that checks for AI hallucinations. Across OCBC, AI is now embedded in more than six million operational decisions daily.

Moving the Risk Conversation Upstream

If source-of-wealth is the on-the-shelf win, pre-qualification is the next layer. Chiang pushed the argument in the same direction as the Hubbis Compliance & Regulatory Forum Singapore 2025 panel agenda, which listed “quicker on-boarding with digital tools” and the challenges of “expedited on-boarding” among its live discussion topics. The Bank of Singapore release itself notes that SOWA “improves account opening efficiency, an area that an industry working group has been set up to look into.”

The logic is to move the risk conversation before business-development momentum has made it commercially and emotionally difficult. Early screening against the bank’s risk appetite gives relationship managers a clear signal on whether a prospect is worth the multi-month chase, and it gives compliance reviewers a head start on the cases that will eventually land on their desks. The goal is not to remove judgement from compliance, it is to move the judgement upstream, where it costs less.

Holistic Risk Across the Bank, Not Just the Compliance Function

Chiang’s definition of “holistic” runs wider than the compliance department. Criminal syndicates know that banks divide responsibility across retail, corporate and private banking lines, and they exploit the seams. The IMTF topic list captures the same instinct at industry level: breaking down silos between KYC, AML, sanctions, fraud and investigations.

Inside one institution, the same logic applies. A client or a network can touch different product lines while the bank’s internal data segmentation makes the pattern harder to see. The MAS enforcement action on the 2023 case found that eight of nine penalised institutions had failed to adequately review transactions their own monitoring systems had flagged as suspicious. That is a cross-system failure, not a single-team failure. Chiang’s argument is that AI has to connect anti-money-laundering intelligence across divisions, not only across compliance disciplines.

AI as a Risk-Culture Lever, Not Just an Efficiency Tool

Chiang’s most distinctive claim is that AI’s deepest role is cultural, not operational. The framing appears in his own writing on the December 2025 Hubbis Compliance & Regulatory Forum in Singapore, where he laid out the position in plain terms: “Use AI to shape RM behavior. Use AI to drive risk culture change. Imagine everyone behavior is shaped to adopt the Bank best practices, complementing our AML training.”

In practice, that means turning each AI-assisted case into a teaching surface. Relationship managers working with the Source of Wealth Assistant are not just producing a report faster, they are seeing how a well-formed source-of-wealth narrative reads against the actual client information in front of them, where the data does not yet support the conclusion, and what a complete assessment looks like.

Three culture levers sit underneath that claim:

• Consistent first drafts. SOWA gives every relationship manager the same starting point, regardless of experience, reducing the variability that Chiang calls out as the human factor’s biggest risk.
• Governed autonomy. The OCBC Group Model Risk Governance framework and the centralised hallucination guard-rails define what the AI can and cannot do, leaving relationship-manager accountability untouched.
• Front-office translation. As Kam Chin Wong puts it, AI lets RMs “shift their focus from manual documentation to meaningful client engagement and risk assessment,” which makes risk standards tangible in the day job rather than abstract policy text.

Chiang is clear on the boundary. “While AI can help to speed up processes, human review remains critical,” he told the Straits Times. The model does not replace the relationship manager; it makes the relationship manager’s judgement more legible and more consistent.

Why the Regulatory Clock Is Ticking: The 2023 Laundering Case

The urgency is regulatory as much as commercial. In August 2023, Singapore authorities seized more than S$3 billion in illicit assets in the largest money-laundering case in the country’s history. On 4 July 2025, MAS concluded its supervisory examinations with composition penalties totalling S$27.45 million against nine financial institutions for breaches of AML/CFT requirements tied to the case.

MAS identified shortcomings in four areas: customer risk assessment, source-of-wealth corroboration, transaction monitoring and post-Suspicious Transaction Report follow-up. Source-of-wealth corroboration was a breach finding for all nine institutions.

Bank of Singapore is not on that list. The same regulatory backdrop, though, is the reason agentic AI for source-of-wealth matters now: the gaps MAS found in 2025 are exactly the gaps SOWA is built to close. MAS has separately proposed AI Risk Management guidelines for all financial institutions, the regulator’s consultation paper setting out supervisory expectations on how institutions should govern, manage and mitigate AI-specific risks. That proposal, not a final rule, is the next milestone the industry is watching.

Frequently Asked Questions

What is SOWA at Bank of Singapore?

SOWA stands for Source of Wealth Assistant, an agentic AI tool launched on 10 October 2025. It drafts source-of-wealth reports for relationship managers, compressing a typical 10-day manual process to as little as one hour and validating client information against OCBC and Bank of Singapore databases.

Who is Kelvin Chiang?

Chiang is Head of Platform & Analytics at Bank of Singapore, OCBC’s private banking arm, with more than 25 years of experience across UBS, Merrill Lynch, Standard Chartered, DBS and Bank of Singapore. He is 52 and leads the financial-crime-compliance analytics function.

What did MAS penalise banks for after the 2023 money-laundering case?

On 4 July 2025, MAS imposed composition penalties totalling S$27.45 million on nine financial institutions for AML/CFT breaches related to the August 2023 case. The named shortcomings were customer risk assessment, source-of-wealth corroboration, transaction monitoring and post-Suspicious Transaction Report follow-up.

What does “agentic AI” mean in this context?

Bank of Singapore defines agentic AI by contrast with generative AI: where generative AI reacts to prompts, agentic systems pursue goals with intelligent initiative, coordinate tools across multiple platforms, adapt dynamically in real time, and continuously refine their performance through memory and learning.

How does AI help with risk culture rather than just efficiency?

Chiang’s argument is that AI turns each relationship-manager case into a teaching surface. The tool produces a consistent first draft, the OCBC governance framework applies the guard-rails, and relationship managers retain accountability for the final review. Culture change comes from the workflow, not from a separate training module.