Undisclosed Global Bank

For one of the world's largest financial institutions, loan processing had become a significant operational bottleneck. The manual review workflow required loan officers to gather and reconcile data from multiple disparate sources — applicant financial records, credit histories, and supporting documentation — before any creditworthiness assessment could begin. Incomplete or inaccurate customer submissions compounded these delays, adding rework loops that stretched timelines across multiple days. Simultaneously, the bank operated under stringent regulatory requirements, demanding rigorous fraud detection protocols and strict data privacy safeguards on all customer information. The combined weight of these constraints made the existing process costly, error-prone, and increasingly difficult to scale to meet growing loan application volumes.

KPMG in the US designed and deployed a customized Generative AI solution built around large language models (LLMs) trained exclusively within the bank's secure environment — ensuring sensitive customer data never left institutional control. The solution used KPMG Ignite, the firm's cloud-based machine learning and data science platform, to develop the extraction model and scale document classification across the multiple data types encountered in loan applications. Critically, KPMG banking sector professionals provided domain-specific input to reinforce and fine-tune the model to the bank's credit policies, operating procedures, and compliance requirements. This human-in-the-loop approach during model development was central to achieving the accuracy thresholds required in a regulated lending context, and the solution was deployed with full integration into the bank's existing loan review workflows.

The Gen AI implementation delivered a dramatic reduction in loan processing time — from multiple days to under one hour — enabling loan officers to make faster, more accurate decisions without sacrificing underwriting rigor. Customers experienced a materially improved experience as approvals accelerated significantly.

Key outcomes:

• Processing time: Reduced from multiple days to under 1 hour
• Operational costs: Reduced through automation of manual document extraction tasks
• Throughput capacity: Increased ability to handle larger application volumes without proportional staffing increases
• Strategic impact: The deployment established a foundation for broader Gen AI adoption across the bank's wider operations

• LLMs trained within a bank's secure, on-premise environment can automate document extraction at scale without creating data privacy or regulatory compliance exposure.
• Fine-tuning AI models with input from banking domain experts — not data scientists alone — is essential for accuracy in regulated lending workflows.
• Processing gains of this magnitude (days to under one hour) directly improve customer satisfaction and competitive positioning in retail banking.
• A successful Gen AI pilot in a high-stakes workflow like credit underwriting builds institutional confidence and creates momentum for broader AI adoption across operations.
• Cloud-based ML platforms enable rapid model development and scaling without requiring extensive in-house AI infrastructure.