↓ Skip to main content
SEA Weekly
SEA Weekly

Southeast Asia

ASEAN AI Brief: What's Driving AI Adoption in ASEAN Trade Finance as Working Capital Stress Tests Traditional Banking Models in Q3

AI adoption in ASEAN trade finance is accelerating at the fraud and compliance layer — but the $2.5 trillion gap is unchanged because AI cannot yet reach the SME tier where the data infrastructure required for credit models barely exists.

Here is a number that gets cited in every fintech conference in Singapore: 84% of banks surveyed in ADB’s 9th Global Trade Finance Gap Survey now use AI for fraud prevention and risk analysis. The number sounds like progress. The sentence immediately after it is the one that should be quoted instead: the global trade finance gap remains at $2.5 trillion, unchanged from two years earlier.

A towering glass financial institution illuminated by sophisticated data flows and AI processing indicators at its upper levels, while at street level below, small-scale factory workers and traders stand in shadow, their hands holding paper documents and manual ledgers — the digital infrastructure visibly failing to reach them
AI is already operating in ASEAN’s trade finance system. The question is which tier it reaches — and which tier it leaves behind.

The two facts — high AI adoption rate, unchanged structural gap — are not a paradox. They are a precise map of where AI is operating and where it is not. That map matters more in Q3 2026 than it has at any point in recent memory, because working capital stress is running across ASEAN’s manufacturing and export sector at exactly the moment when the case for AI-enabled trade finance sounds most compelling. The compelling case and the structural reality are not the same thing.

Where AI actually operates in ASEAN trade finance
#

The 84% figure from ADB’s January 2026 survey is accurate. It describes what the technology is doing: AI models running across transaction flows at large commercial banks and development finance institutions, identifying anomalous patterns associated with fraud and money laundering, flagging documentation discrepancies in letters of credit, and generating risk scores on counterparty exposure. These are the applications that work because they operate on clean, structured, high-volume data that large banks already had before AI arrived.

What that 84% number does not describe is AI expanding the frontier of credit access. It describes AI making the existing frontier more efficient. DBS, Standard Chartered, HSBC, OCBC — the banks that between them executed or are pursuing over $4.5 billion in significant risk transfers in recent weeks to free up regulatory capital for trade finance — are using AI to process their existing deal flows faster and more accurately. They are not using it to approve credit to Vietnamese garment manufacturers who do not have audited financial statements or to Indonesian SMEs whose transaction histories exist primarily in cash and informal supplier ledgers.

The 57% of banks telling ADB they are “exploring how AI can expand financing capacity” is the more interesting number. Exploring is not deploying. And the gap between exploring and deploying is not a regulatory or technology barrier — it is a data infrastructure barrier.

Q3 stress is testing this structure in real time
#

The working capital pressure building across ASEAN in Q3 2026 has a specific anatomy. As Chloe Tan documented on July 8, it operates through three simultaneous channels: rising freight costs extending the financing gap on every shipment, buyer-seller tension lengthening effective cash conversion cycles, and broad-based credit tightening that is hitting SMEs hardest even as ADB’s data appears to show convergence with large-corporate rejection rates.

That apparent convergence — SME rejection rate at 41%, large-corporate rejection rate at 40% — is the data point in the ADB survey most worth examining carefully. ADB noted it “requires more research,” which in development bank language signals unease with the result. The most plausible reading is not that AI has improved SME access to trade finance but that tighter global conditions have worsened large-corporate access. Progress by convergence from the wrong direction.

The firms absorbing Q3 stress at full exposure are precisely the ones that AI credit models cannot yet serve: Vietnamese MSME exporters running 90-day receivable cycles without credit bureau coverage; Indonesian manufacturers whose input costs just hit the highest level since September 2013 and whose loan applications are being evaluated by OJK’s “persistently weak” MSME lending infrastructure; Philippine inter-island logistics operators bridging food distribution costs without the EDI integrations that any machine learning rate model requires.

ADB’s response to this structural gap in Q3 has been instructive: a $721 million multilateral facility through HDBank specifically structured to route credit toward Vietnamese MSMEs, mobilising 29 commercial banks to reach firms that the commercial market was failing to serve. The fact that Vietnam — which has 97% of its registered businesses in the MSME tier, representing 40% of GDP — required a seven-hundred-million-dollar multilateral construction project to channel credit to that tier tells you everything about where AI-enabled commercial trade finance has and has not reached.

MAS SAFR and the agentic finance map
#

The most significant regulatory development in ASEAN AI finance this week was not a trade finance announcement. On July 3, MAS published the Safeguards for Agentic Finance at Runtime (SAFR) framework — an industry white paper developed under its BuildFin.ai initiative with leading financial institutions and fintechs. SAFR defines governance checkpoints for AI agents in financial services: how their actions are authorised, how human oversight is activated, and what is recorded at the point of every decision.

The use cases MAS and its industry partners chose to pilot under SAFR are a precise indicator of where the data infrastructure supports agentic AI deployment:

  • Agent-assisted payments and treasury operations, where autonomous agents execute routine transactions within predefined mandates
  • Wealth management and advisory workflows, where AI agents review documents and generate structured assessments
  • Client engagement, where AI agents generate client insights and draft materials within approved content boundaries

Notice what is absent: invoice financing for SME exporters, working capital credit assessment using alternative data, supply chain finance for manufacturing tier suppliers. These are not regulatory oversights. They are absent because the data infrastructure required to run responsible agentic AI in those use cases — clean transaction histories, structured counterparty data, reliable documentation flows — does not exist at scale in the ASEAN SME manufacturing and export sector that accounts for the bulk of the $2.5 trillion gap.

MAS’s newly established Future of Finance Institute will support pilots and sandbox experimentation for SAFR-aligned solutions. This is the right institutional infrastructure. But it is infrastructure at the proof-of-concept stage in July 2026, not at the deployment scale that Q3 working capital stress requires.

The data problem that AI adoption figures obscure
#

My July 5 analysis of AI logistics tools made a similar structural argument about freight: the AI tools that manage freight cost volatility are effective for the large operators that don’t need the most help, and they’re out of reach for the SME exporters who do. The argument in trade finance is structurally identical.

The barrier in both cases is not the quality of the AI models. Blue Yonder’s supply chain control towers work. The fraud detection models running in DBS and Standard Chartered work. The alternative-data credit scoring platforms being piloted in Singapore’s fintech ecosystem work. The barrier is the underlying data infrastructure those models require to function — and that infrastructure is absent in exactly the parts of ASEAN’s trade and logistics economy where the structural gap lives.

For an AI credit model to assess a Vietnamese garment exporter’s working capital request, it needs: two to four years of structured transaction history, documented buyer-supplier relationships, verified export documentation, and receivables data in a format that machine learning pipelines can process. A significant share of Vietnam’s 800,000-plus SME enterprises do not have all four. They do not lack creditworthiness — they lack the data trail that AI credit models require to recognise creditworthiness. The ADB’s multilateral intervention is not a substitute for AI-enabled credit; it is a patch for the period before AI-enabled credit can function in this tier.

The 2030 test
#

ADB’s own analysis identifies digitalizing trade documentation by 2030 as critical to closing the global trade finance gap. ADB’s July 2026 growth downgrade — developing Asia at 4.9% for 2026, regional inflation at 4.3% — sets a more difficult environment in which to close that gap than the April forecasts implied. Working capital stress compresses the time horizon for ASEAN manufacturers and exporters. Institutions that cannot access adequate trade finance in Q3 2026 do not survive to participate in a 2030 digital infrastructure rollout.

The Q3 pressure cycle will pass. Freight rates will normalise, Indonesia’s PMI will recover, Vietnam’s H1 trade deficit will convert into export revenue in Q3 and Q4 as ordered production flows through. But the structural argument — that AI adoption in ASEAN trade finance is concentrated at the top of the capital stack and the gap lives at the bottom — will not resolve with the rate cycle.

The honest measure of progress in ASEAN AI trade finance adoption is not what percentage of banks have deployed AI. It is whether the AI being deployed is expanding the frontier of credit access into the tier where the $2.5 trillion gap actually lives. On that measure, Q3 2026 is an important stress test — and the results are still coming in.

References: