How a major UK bank cut mortgage offer times with agentic AI

This bank faced hurdles that were inflating costs, causing rework, and creating a frustrating experience for mortgage brokers.

Mortgage processing relied heavily on manual activities across document review, application setup, underwriting preparation, and offer issuance. Case packaging required extensive manual effort and resulted in rework rates of up to 60%. Underwriters spent roughly half of their time on administrative tasks, including reviewing supporting evidence, validating customer information, completing affordability calculations, and requesting missing documentation from brokers. These inefficiencies increased turnaround times and created friction for both employees and brokers.

Following an AI value assessment of the mortgage journey, we identified multiple points where manual document handling and data validation were slowing the process. Rather than introducing isolated automation, we designed a coordinated network of AI agents that work together across the mortgage workflow. With our last mile execution into the flow of work i.e. connecting documents, data, processes, and decisions so AI could create measurable value in day-to-day mortgage operations.

We built four customized AI agents that span different stages of the mortgage offer process:

1. Standardization agent: Standardizes mortgage application documents

2. Completeness and classification agent: Checks document completeness and classifies documents, so nothing falls through the cracks

3. Extraction and verification agent: Extracts and validates application information

4. Income calculator/budget planner agent: Supports affordability assessment and budget planning and prefills the affordability calculator

The agents operate within an orchestrated workflow managed through AWS Step Functions, with AWS Lambda providing serverless processing. Once an agent completes its task, the application automatically progresses to the next processing queue, triggering the subsequent agent. This sequential workflow enables the agents to collaborate without requiring a separate orchestration agent. Mortgage applications and supporting documents enter the AWS environment through APIs, are processed by the AI agents, and the resulting outputs are returned to the mortgage workflow for underwriter review.

Built securely in the cloud using generative AI, the solution removes the heavy lifting from the early stages of a mortgage application.

The solution uses:

• Amazon Textract for document extraction

• OpenAI models for AI-driven document understanding and processing

• AWS Lambda for serverless processing

• AWS Step Functions for workflow orchestration

• Amazon API Gateway as the API connectivity layer, enabling secure data exchange between mortgage systems and AI services

• Amazon S3 as the primary storage repository for mortgage applications, supporting documents, and processing outputs

• Amazon DynamoDB as the database layer for application data storage and workflow tracking