Read More

Executive Summary

A commercial bank in Indonesia initiated an Intelligent Document Processing (IDP) program to improve efficiency, accuracy, and governance in credit assessment workflows, particularly for Bank Account Statement (BAS) processing. The Bank Statement Analyzer (BSA) solution was implemented to automate data extraction, validation, and document forensic checks while reducing reliance on manual data entry.

The platform leverages AI-powered OCR to process structured and unstructured BAS documents and generate standardized outputs aligned with the bank’s reporting requirements. The system operates in a dedicated single-tenant environment deployed in AWS Jakarta to support data residency, security, and enterprise operational requirements.

Challenge

The bank faced operational and governance constraints in BAS processing that impacted speed, consistency, and scalability:

1. Manual extraction and review bottlenecks
   Credit assessment teams spent significant effort extracting transactions from diverse statement formats, slowing turnaround time and increasing operational workload.
2. Data quality inconsistency and fraud exposure
   Variations in statement layouts and document types increased the risk of human error and reduced consistency in fraud detection and screening.
3. Limited auditability and governance controls
   Credit operations require strong traceability, who uploaded, reviewed, edited, exported, and approved outputs, supported by bank-grade access control and logging.
4. Scalability and performance requirements
   Higher document volumes require bulk submission, asynchronous processing, and monitoring to maintain predictable turnaround times and productivity.

Solution

The Bank Statement Analyzer (BSA) was implemented to automate BAS extraction, validation, and fraud screening while establishing enterprise-grade security, governance, and scalability controls.

AI-powered BAS Extraction and Validation

• Supports multiple input formats including PDF, TIFF, PNG, scanned documents, and password-protected files
• Processes both structured and unstructured bank statements
• Achieved 95%+ extraction accuracy based on the bank sample benchmarking
• Supports bulk upload capability of up to 30 files per batch
• Provides human-in-the-loop correction with full audit trail for manual edits

Fraud Detection and Document Forensic (Baseline)

• Performs document forensic checks such as font consistency analysis and transaction pattern flagging to support fraud detection and reviewer assessment.
• Validates document integrity indicators such as metadata and tampering signals
• Produces fraud detection outputs to support reviewer assessment

Standardized Reporting Outputs

• Generates Excel outputs aligned with the bank reporting templates
• Produces structured data suitable for downstream processing and analytics

Enterprise Security and Governance Controls

• Implements Role-Based Access Control (RBAC) with centralized authentication via Active Directory/LDAP
• Captures comprehensive audit trails covering login/logout, document processing, manual edits, exports, and user maintenance
• Enforces business-unit level access segregation and monitoring
• Implements automated backup and retention strategy aligned with internal IT and Cyber Security policies
  

High-Level Architecture

The solution is deployed in AWS Jakarta (ap-southeast-3) using a layered architecture designed for security, scalability, and high availability.

Access Layer

• CloudFront and Application Load Balancer manage external access
• Optional secure connectivity via Site-to-Site VPN for the bank network access
• Internet Gateway acts as the controlled entry point into the AWS VPC

Application Layer

• Runs on Amazon EKS (Kubernetes) within private subnets
• Includes frontend application, backend APIs, and background workers
• Queue-based asynchronous processing supports large document workloads
• GPU-based AI processing services are deployed as containerized workloads within EKS, utilizing LiteLLM and vLLM as the AI gateway and LLM processing engine, to support IDP, document understanding, and semantic processing tasks with Qwen 2.5 7B Instruct and 32B Instruct models.

Data Layer

• Amazon RDS PostgreSQL with Multi-AZ failover as the primary database
• Amazon S3 for document and output storage
• Amazon SQS for message queuing and job orchestration

Security and Availability

• Multi-layer network isolation across public, application, and data subnets
• Encryption in transit (TLS/HTTPS) and at rest (KMS)
• Multi-AZ deployment with auto-failover and auto-scaling capabilities
• Automated backup processes supporting business continuity objectives

Outcome

• Faster BAS Processing and Higher Throughput: Automated OCR processing enables bulk ingestion of up to 30 bank statement files per batch, significantly reducing manual document handling and accelerating BAS analysis cycles.
• High Data Extraction Accuracy: The AI-powered OCR engine achieves ≥ 95% extraction accuracy on benchmarked CIMB bank statement samples, reducing manual correction effort and improving reliability of transaction-level outputs.
• Early Fraud Detection and Risk Identification: Automated document forensic analysis (e.g., font consistency checks against bank template libraries) enables early detection of potentially manipulated bank statements before financial analysis processing
• Strengthened Security and Governance: The system enforces role-based access control (RBAC), comprehensive audit logging, and automatic session termination after 15 minutes of inactivity, ensuring compliance with internal IT security and access governance policies.
• Reliable and Resilient Data Platform: The platform implements automated backups and disaster recovery controls with daily backup execution, RPO of 4 hours, and RTO of 2 hours, ensuring business continuity and rapid recovery in failure scenarios.
• Scalable Enterprise-Ready Architecture: Deployment on AWS Jakarta with Multi-AZ infrastructure, auto-scaling compute workers, and event-driven processing allows the system to scale dynamically with document volumes and user growth.

About Mastersystem Infotama

PT Mastersystem Infotama Tbk (MSTI) has developed into one of the leading ICT infrastructure providers and dominates not only the enterprise banking market in Indonesia, but also the oil and gas, enterprise and telecommunication industry. MSTI is an Amazon AWS Advanced Tier Services Partner, with this accreditation allows MSTI to collaborate even more closely with Amazon AWS and continue to drive customers toward success leveraging the cloud.