Automated Report Generation & BI Solution
May 28, 2026
About Customer
Exovance Global is a global professional services and outsourcing company that delivers specialized solutions across accounting, finance, and business operations. The company focuses on helping organizations improve efficiency, reduce operational costs, and scale their back-office functions through expert-driven services.
Exovance Global provides a wide range of offerings, including accounting services for firms, payroll management, bookkeeping, year-end accounts preparation, paraplanning, and mortgage processing services. Their solutions are tailored to meet the specific needs of accounting firms and financial service providers, enabling them to streamline workflows and enhance productivity.
By combining domain expertise with process optimization and technology-enabled delivery models, Exovance Global supports clients in achieving operational excellence while maintaining high standards of accuracy and compliance. Their service approach emphasizes flexibility, scalability, and cost-effectiveness, making them a trusted partner for firms looking to outsource critical business functions.
Business Problem
Like many fast-growing lending institutions, Loan Intel’s customers soughtto fulfill two fundamental business needs: operational efficiency and consistency at scale.
Mortgage underwriters were spending the majority of their time manually extracting data from financial documents — payslips, bank statements, loan agreements, and tax certificates. This manual process was slow, error-prone, and impossible to scale as application volumes grew.
Self-employed applicants were particularly challenging. Tax forms and Form 11 documents require specialist knowledge to interpret, and different underwriters applied different judgment to similar cases —leading to inconsistent eligibility decisions.
Without automation, the institution risked losing customers to faster competitors, increasing operational costs with every new hire, and facing compliance risks from inconsistent decision-making across the underwriting team.
Solution
To support mortgage lenders in achieving automated and consistent under writing, Loan Intel implemented an AI pipeline on AWS that uses Amazon Bedrock (Nova Pro) and GCP Gemini with a 7-Lambdasequential processing architecture.
The first component, the Income Analysis module, uses AWS Textract for OCR and Amazon Bedrock to extract salary data from payslips and salary certificates. It validates consistency between the two documents, detects bonus payments, and handles self-employed income and Form 11 tax documents.
The second component, the Expense & Loan Analysis module, processes bank statements using GCP Gemini 2.5 Pro to identify recurring transactions, flag suspicious activity, and extract EMI obligations from loan statements. It cross-validates expense dataagainst income data to ensure accuracy.
The third component, the Summary Analysis module,combines all extracted data to calculate the debt-to-income ratio,generate an eligibility score (0–100), and produce a final decision— Approve, Conditional, or Reject — with detailed reasoning andunderwriter recommendations.
The Generative AI pipeline is built on a serverless architectureusing AWS Lambda that provides scalability and cost efficiency.The user interface is deployed on AWS Amplify, allowingunderwriters to upload documents, track processing in real time,manage queries, and view eligibility results. The entire system isdeployed via AWS SAM with a GitHub Actions CI/CD pipeline.
Architecture Overview
Results
Thanks to the automated pipeline, mortgage applications thatpreviously required 2–4 hours of manual document review are nowprocessed in approximately 2.5 minutes — a 98% reduction inprocessing time.
The system handles unlimited concurrent applications with noadditional cost per application, enabling the institution to scale withouthiring additional underwriters. Underwriters now focus exclusively onquery resolution and final decisions rather than data extraction.
The cross-validation between income, loan, and expense data hasimproved accuracy — salary mismatches between payslips andcertificates are automatically flagged, and suspicious transactions areidentified without manual review. Confidence scores on every decisiongive underwriters clear guidance on when to approve automaticallyand when to review manually.
Additionally, the per-file OCR caching system ensures that onreanalysis, only newly uploaded documents are re-processed —reducing repeat processing costs by up to 75% compared to full re-extraction.
