1 of 4

Loan Origination Process

Agentic Workflow

Loan Application Received

Robot

Validate Application Completeness

Robot

Request Additional Information

Robot

Check Application Completeness

Robot

Update LOS, Additional Checks

Robot

Parallel Validation checks

Robot

Risk Calculations: DTI, LTV, Affordability, etc.

Agent

Eligibility Analysis

Human + Autopilot

Review Escalation

Agent does the time-consuming back and forth, people help review

Agent

Investigation of policies & system data

Robot

Update LOS, Generate Contract

Robot

Notify Applicant

Loan Application Denied

Rejected

Approved

Robot

Read Application Documents (IXP)

Completed

2 of 4

Agent example �Loan Origination Process

think

Act

Agent

Agent	Specification
Persona/Role	“Act as a Loan Originator responsible for evaluating loan applications against policy”
Role Activities	“Your activities include reading information from the LOS system, invoking robots to evaluate the information, and sending notifications to human reviewers or the customer directly about the decisions taken.”
Knowledge Context	Bank Policy documents, risk assessment models, bank balance sheets, regulator policy documents
Quality Measure or KPI	Time to provide decision to customer, time spent making loan decisions
Input Context	Loan application, current bank rates, loan applicant auxiliary data (credit report, IDs, income verification)
Output Instructions	“Provide either an approved or declined decision to the customer with justification, or escalate for human review”

Robotic feature	Requirements
LOS (Loan Origination System), Credit Report, and other services	RPA to retrieve applicant and application loan data to process in auxiliary systems and process data.
UiPath API Integrations	Data Extraction from documents, Document storage, and sending and receiving communication with customers.
UiPath Document Understanding Automations	RPA extracts data from customers document for verifications (income, and others) and eligibility.
Email Automation and Communications Mining	RPAs Extract data from customer communication, add data to applications, and send information to agent to process.
Data Validation and Exception Handling with Action Center	RPAs ensuring accuracy of application data before it is sent and flag any discrepancies for human review.
Reporting Tools (e.g., UiPath Insights)	Dashboards for tracking the process of the application, monitoring application status, and providing data on risk and profitability performance.

3 of 4

Loan Eligibility Analysis Agent: An agent to intelligently analyze all relevant information of a potential loan, ensure compliance with company and regulator policy, utilize context grounding for nuanced evaluation of loan details, and provide either a final approval or an escalation for a human loan originator.

Problem Statement: Speed the process from application to closing, ensure risk mitigation, increase low risk application approval and reduce manual effort required.

Analysis / Changes: Communication mining to speed up communication between customer and us, document data extraction and application completeness automation.

Analysis (AS-IS): Too much time reaching out to the customers and waiting for additional information/documentation. Too many checks that are missing before first review.

Changes (TO-BE): Automate the Communication process between customer and us for application data completion and follow-up. Automate the validation and policy/guidelines automation with an agent. Escalation for exceptions and final approval.

Results / Benefits:

Significant reduction in manual review workload for simple unsecured loans.
Faster time-to-decision for low-risk applicants.
Higher straight-through processing (STP) rate with minimal human oversight.
Better customer experience due to quicker feedback and decisioning.

Category	Before	After	% Improvement
Decision Turnaround Time	2–3 business days	Same-day or instant decisions for most applications	80%
Manual Intervention Rate	100% of applications required human review	40% of applications require human review	60%
Monthly Processing Capacity	~1,500 loans with 7 FTEs	>3,000 loans with same headcount	100%
Policy Adherence Consistency	Varied by reviewer	Standardized through automated validation rules	High Consistency

Process Diagram:

Results / Benefits:

Automated eligibility checks against current policy and risk models.
Clear audit trail for all automated decisions.
Reduced human error and improved consistency in loan evaluations.
Scalable solution enabling increased loan volume without increasing headcount.
Automated eligibility checks against current policy and risk models.
Clear audit trail for all automated decisions.
Reduced human error and improved consistency in loan evaluations.
Scalable solution enabling increased loan volume without increasing headcount.

Application Data Validation

Get Auxiliary Date (Credit Report)

Calculate DTI/LTV…

Eligibility Agent

Human-in-the-loop escalation

Approved?

Send Decline Letter

Generate Contract

Notify Customer

Declined

Approved

New Application

4 of 4

Loan Eligibility Analysis Agent

Agent Tools:

Read Application Documents (IXP): collect new files from Inbox, get PDF/Images attachments for Extraction
LOS Communicator: Read data from LOS for evaluation and update data in LOS from extraction and evaluation results
Criteria evaluation Robot: Handle structured rule validation, evaluate LTV, DTI and risk against company policy. Provide pass or fail argument back to agent.
Escalation UiPath App: Custom App for humans to evaluate information already gathered and evaluated by bot.
Applicant Notification Bot: Take all collected data and decision from agent or human and notify loan applicant of approval or denial with appropriate details.

18,000+

application per year

80%

improvement in time savings

12 weeks

of development

60%

STP without human involvement

Business Benefits:

Standardization: Development split by loan segment and type
Stability: Manual, repetitive, and time-consuming processing tasks eliminated.
Scalability: Increase loan volume without increase in overhead.
Leverage Modules: Expand to other areas of loan origination.
Productivity: Loan Processing and Approval team needs 7FTE’s to process 1500/month with a lot of manual work.
Accuracy: ML Model at 87,1% Accuracy and retraining aiming to 92% target

Lessons Learned

Don't try to do too much

Development split by loan segment and type
Built and trained the data model for each documents received

Have process transparency

Controls and Warnings in the Process (QA & QC Reports)
Real-time model and process performance metrics

Have a failsafe – pass the transaction back to the team
Open ended resource allocation

The best improvement ideas will appear once the team embeds the tool into everyday use

System Prompt:�You are an intelligent loan evaluation agent responsible for assessing applications for simple unsecured loans. You read from and write to the Loan Origination System (LOS), and your decisions are grounded in the latest company policy documents and frequently updated risk acceptance models. Evaluate each application based on these criteria. Approve or reject automatically when confident. Escalate to a human only when risk thresholds, policy ambiguities, or edge cases are detected.