Case study
70% Faster Loan Processing with an AI-Powered Banking Knowledge Management System
- Team size: 20
- Development time: 8 months
Explore more case study
Background
Deloitte’s 2026 banking outlook says many banks are still held back by brittle, fragmented data foundations, and that AI programs often remain isolated projects rather than scaled operating models. McKinsey’s 2026 research also notes that only about 1/3 of organisations report high maturity in strategy, governance, and agentic AI governance, which shows how difficult it still is to operationalise AI responsibly at scale.
For our client, a Vietnam-based commercial bank modernising lending operations, these challenges appeared in two critical areas.
- First, the bank lacked a centralised banking knowledge management system. Regulations, policy manuals, product sheets, pricing documents, promotion leaflets, and training materials were stored across disconnected repositories. Employees relied on exact keyword search and manual knowledge lookup, making it difficult to quickly retrieve accurate information or provide consistent advisory support to customers.
- Second, the loan approval process remained highly manual and fragmented. Analysts and underwriters spent significant time reviewing documents, extracting information, validating compliance requirements, and assessing risks across multiple systems. This slowed down loan processing and increased operational overhead.
To address both challenges, GEM designed a scalable AI-driven solution combining a banking knowledge management system with an AI-powered loan decision support architecture.
Challenges
Fragmented banking knowledge and inefficient information retrieval
- Banking regulations, policies, product sheets, and training documents were distributed across disconnected systems
- The legacy search engine relied on exact keyword matching instead of semantic search capabilities
- Information retrieval was slow, inconsistent, and difficult to scale across lending operations
- Sales teams struggled to consistently identify the correct credit segment and provide accurate advisory support
- Fragmented knowledge access increased customer experience and compliance risks
Manual and time-consuming loan decision workflows
- Manual document handling and information extraction
- Analysts and underwriters spent significant time validating risk and compliance requirements
- Customer and lending data had to be reviewed across multiple disconnected workflows
- Manual review processes reduced underwriting efficiency and slowed loan approvals
- High operational effort limited the scalability of lending operations
Solution
The solution was delivered through two connected workstreams.
Workstream 1: Banking Knowledge Management System
GEM built a centralised banking knowledge management system that consolidates regulations, policies, product sheets, pricing documents, and training materials into a unified enterprise repository.
The platform supports semantic and intent-based search, multi-document intelligence, and an AI sales advisor assistant that provides real-time guidance on credit segments, banking products, and policy information.
Workstream 2: Loan Decision Multi-Agent Support System
GEM implemented an AI-powered loan decision support system using a multi-agent processing architecture for document ingestion, data extraction, risk analysis, compliance validation, and decision support.
The solution combines automated document understanding, intelligent loan assessment, and human-in-the-loop review workflows so analysts and underwriters can validate AI recommendations before final approval.
Tech stack
- OCR
- MCP
- API interface
- LOS
- National CIC
Output
- Delivered a centralised banking knowledge management system integrating regulations, policy manuals, product sheets, pricing documents, promotion leaflets, and training materials
- Enabled semantic and intent-based enterprise search across banking knowledge assets
- Built an AI sales advisor assistant supporting real-time guidance on credit segments, banking products, and policy recommendations
- Implemented an AI-powered loan decision support system with automated document ingestion and intelligent data extraction
- Enabled AI-powered risk analysis, compliance validation, and recommendation workflows for lending operations
- Integrated human-in-the-loop review workflows for analyst and underwriter validation before final approval
- Established vector search, ontology-based knowledge mapping, and enterprise graph capabilities for contextual knowledge retrieval
- Connected lending workflows with LOS and National CIC integration interfaces for operational decision support
Impacts
Operational efficiency
- Loan application processing became 70% faster
- Manual data handling was reduced by 60%
- Analysts spent significantly less time on manual document review and information retrieval
- Loan processing workflows became more standardised and scalable across lending operations
Business transformation
- Faster loan approvals improved responsiveness for customers and reduced operational bottlenecks
- Reduced manual processing effort helped lower operational costs per application
- More consistent advisory and underwriting processes reduced compliance and operational risk
- The bank established a stronger foundation for scalable AI-driven lending operations and future digital banking initiatives
Closing remarks
The banking knowledge management system and AI-powered loan decision system enabled the client to modernise lending operations with faster and more scalable decision-making.
Discover another AI transformation success story: Intelligent shipper and carrier matching platform
