CS-03
Pan-ASEAN Bank — AI Credit Underwriting at Scale
A top-5 Pan-ASEAN regional bank operating across Singapore, Malaysia, Indonesia, Thailand, Vietnam, and the Philippines faced a critical challenge: manual credit underwriting was costing the bank $180M in foregone SME credit opportunity while exposing it to rising non-performing loans. Anicalls deployed its Credit Intelligence Engine™ across all six markets — delivering 90% of credit decisions within four hours and reducing the NPL ratio by 65% in 18 months.
90%Decisions in 4 Hours
65%NPL Ratio Reduction
$180MNew SME Credit Book
55%Underwriting Cost Down
Business Challenge
Why Manual Underwriting Was Failing Across Six ASEAN Markets
14–21 Day Decision Cycle
Manual credit assessment required bureau pulls from 6 different national credit bureaus, manual cash flow reconstruction from bank statements, and up to 8 credit officer touch-points per application. SME applicants dropped off at 40% — choosing competitors or informal lenders rather than waiting 3 weeks for a decision.
Rising NPL Ratio
The bank's SME NPL ratio had climbed to 3.2% — well above the ASEAN regional average of 1.8% — driven by inconsistent underwriting quality across markets. Criteria varied between Singapore, Indonesia, and the Philippines, with no centralised model validation. Each market used separate scoring cards that had not been recalibrated since pre-COVID.
Multi-Market Complexity
Six markets meant six regulatory frameworks (MAS, BNM, OJK, BOT, SBV, BSP), six local bureau integrations, and credit documentation in Bahasa Indonesia, Thai, Vietnamese, and Filipino alongside English and Malay. Credit officers lacked tools to assess cross-border SME groups with entities in multiple ASEAN jurisdictions simultaneously.
Credit Officer Capacity Crisis
Each credit officer was managing 800+ active credit files per month — a volume 3× industry benchmarks. 60% of officer time was spent on data gathering and document verification, not credit judgment. Attrition among credit officers reached 28% annually, with institutional credit knowledge walking out the door. The bank was unable to scale the SME credit book without proportionally growing headcount.
Solution Delivered
Anicalls Credit Intelligence Engine™ — ASEAN Multi-Market Deployment
Credit AI
Automated Bureau Aggregation & Scoring
Anicalls integrated with all 6 national credit bureaus simultaneously — CCRIS/CTOS (Malaysia), SLIK (Indonesia), Credit Bureau Singapore, NCB (Thailand), CIC (Vietnam), and CIC (Philippines) — pulling bureau data in real time and synthesising cross-market exposure for group-level SME assessments.
- 6-bureau real-time API integration
- Cross-market SME group consolidation
- Bureau data quality scoring and gap-filling
- Alternative data enrichment (telco, e-commerce, tax)
Decisioning
AI Credit Decisioning Engine
Machine learning credit scoring models trained on 8 years of the bank's own credit history, recalibrated per market with local macroeconomic variables. The engine produces an automated decision recommendation with full explainability output compliant with MAS FEAT principles — enabling credit officers to approve, override, or escalate in minutes rather than days.
- Market-specific ML scoring models (6 variants)
- MAS FEAT explainability compliance
- Automated approve / refer / decline recommendation
- Human-in-loop override with documented audit trail
Compliance
Multi-Regulatory Compliance Layer
The platform enforces per-market regulatory limits, concentration risk rules, and responsible lending requirements automatically. BNM Responsible Financing Guidelines, OJK POJK 35 consumer protection, and MAS Notice 635 are encoded as hard constraints — preventing non-compliant credit decisions from reaching approval queues and reducing regulatory exposure across all markets.
- MAS Notice 635 / BNM RFG / OJK POJK 35 constraints
- Concentration risk monitoring and alerts
- Regulatory reporting automation (6 markets)
- Data localisation enforcement per jurisdiction
AI Workforce Deployment
How the Anicalls AI Credit Team Was Structured
AI Credit Analysts (Tier 1)
24 AI Credit Analyst agents handle the initial application triage — bureau pulls, document verification, cash flow reconstruction from uploaded bank statements, and preliminary credit scoring. Operating 20 hours per day across SGT and WIB time zones, each AI agent processes 200 applications per day — replacing 3 FTE analysts per agent at 55% of the cost.
AI Underwriting Specialists (Tier 2)
12 AI Underwriting Specialist agents handle complex referrals — group-level consolidations, cross-border SME assessments, restructured facilities, and applications requiring alternative data sources. These agents produce a complete credit memorandum with recommendation, supporting rationale, and MAS FEAT explainability report within 45 minutes of referral.
Multilingual Document Processing
Specialist language processing agents handle financial documents in Bahasa Indonesia, Thai, Vietnamese, and Filipino — extracting audited financials, tax certificates, and trade references from non-English source documents. OCR accuracy of 99.1% on structured financial documents, with human review triggered for documents below 95% confidence threshold.
Technologies Used
The Anicalls Technology Stack Behind This Deployment
Core Platform
Credit Intelligence Engine™ v3.1
Anicalls' purpose-built credit decisioning platform — pre-integrated with 40+ APAC credit bureaus, trained on $12B of ASEAN credit history, and configurable per-market credit policy rules. Deployed on-premise at the bank's Singapore data centre with data residency controls enforced per jurisdiction.
- Multi-bureau API orchestration layer
- Market-specific ML model registry (6 models)
- Policy rules engine (hard / soft constraints)
- Credit memorandum auto-generation
AI Models
ASEAN SME Credit Scoring Models
Gradient boosting ensemble models trained per market on the bank's 8-year credit history, enriched with macroeconomic indicators, property price indices, sector health scores, and alternative data signals from telco usage, e-commerce transaction volumes, and tax filing regularity. Models recalibrated quarterly with Gini coefficient monitoring.
- XGBoost / LightGBM ensemble per market
- Alternative data integration (telco, e-commerce, tax)
- Quarterly recalibration with drift detection
- SHAP explainability for MAS FEAT compliance
Document AI
Multilingual Financial Document AI
Computer vision and NLP pipeline for financial document processing in 6 languages — extracting P&L data, balance sheet figures, cash flow statements, and tax filing data from unstructured PDFs, scanned documents, and photos. Integrated with the bank's existing core banking system (Temenos T24) via pre-built API connector.
- 6-language OCR and NLP pipeline
- Structured financial data extraction
- Temenos T24 core banking integration
- Document authenticity verification (fraud signals)
Quantified ROI
The Financial Impact Delivered at 18-Month Mark
$180M New SME Credit Book
Faster decisions and reduced application dropout (from 40% to 8%) enabled the bank to originate $180M in new SME credit over 18 months — credit that would have been lost to competitors or informal lenders. Higher-quality decisioning also improved margin: AI-scored loans averaged 40bps higher yield than manually underwritten equivalents due to better risk-pricing accuracy.
$42M NPL Provision Release
The 65% NPL ratio reduction across the AI-underwritten portfolio — compared to the pre-AI cohort — released $42M in loan loss provisions previously held against the SME book. Improved early warning indicators (30-day arrears detected 60 days earlier than legacy systems) enabled proactive restructuring, converting potential write-offs into performing recoveries.
$28M Annual Cost Reduction
The 55% reduction in underwriting cost — achieved by replacing 3 FTE equivalents per AI agent across 24 analyst agents — delivered $28M in annual operating cost savings. Credit officer headcount was redeployed to complex relationship banking and product cross-selling, increasing revenue-generating activity by 4× per officer FTE. Platform ROI: 520% in Year 1.
Business Outcomes
Strategic Transformation Beyond the Numbers
Market Position
ASEAN SME Credit Leader
The bank moved from #4 to #2 in ASEAN SME lending market share within 18 months of deployment. The 4-hour decision SLA — communicated in marketing materials — became a competitive differentiator in markets where the next-fastest competitor took 5 business days. Net Promoter Score among SME borrowers increased from 28 to 61.
Risk Quality
Best-in-Class Portfolio Quality
AI-underwritten vintages outperformed legacy vintages by 180bps in 12-month NPL rate across all six markets. The AI models detected early stress indicators — sector concentration, related-party exposure, cash flow seasonality — that manual underwriters consistently missed. The bank's credit rating agency (Moody's) cited improved underwriting quality in its stable outlook affirmation.
Scalability
Zero-Headcount Growth Model
Application volume grew 3× in 18 months — from 12,000 to 36,000 monthly applications — with no increase in credit officer headcount. The AI platform scaled elastically to handle volume peaks (end-of-month, pre-holiday SME credit demand surges) without quality degradation. The bank expanded into two new ASEAN markets (Cambodia and Myanmar) using the same platform, with 45-day market onboarding time.
Executive Testimonial
"We had accepted slow credit decisions as an inevitable cost of operating responsibly across multiple ASEAN regulatory environments. Anicalls proved that assumption wrong. Within six months, our credit teams were spending their time on relationship banking — not chasing documents — and our SME customers were getting decisions faster than any competitor in the market. The NPL improvement was unexpected but logical: better data, better decisions, better portfolio. This is the most impactful technology deployment we have undertaken in a decade."
Metrics Dashboard
18-Month Performance Scorecard
4hrsAvg Credit Decision (was 14 days)
90%Applications Auto-Decided
65%NPL Ratio Reduction
$180MNew SME Credit Originated
$42MProvisions Released
55%Underwriting Cost Reduction
8%Application Dropout (was 40%)
520%Platform ROI Year 1
Transform Your ASEAN Credit Operations
See how the Credit Intelligence Engine™ can reduce your decision cycle from days to hours — while improving portfolio quality across every ASEAN market you operate in.