Loan Underwriting Agent
Build an agent that underwrites loan applications — weighing credit, income, and risk against policy, deciding the outcome, and knowing exactly which deals a human underwriter must sign off.
Challenge Overview
This challenge models the underwriting desk of a digital lender, where an agentic underwriting bot turns around a queue of 6 loan applications for a same-day decision.
For each application the agent weighs several factors at once — credit score, debt-to-income, employment history, and default record — then decides: Approve, Refer to underwriter, or Decline, logs the driving reason, and flags any large loan for senior sign-off. It is a practice simulation of numeric risk reasoning — not real financial advice.
Underwriting Policy
Evaluate in order — the first matching rule wins.
1.Prior default on record
Decline2.Credit score below 600
Decline3.Debt-to-income above 45%
Decline4.Score ≥ 720 · DTI ≤ 35% · 2+ yrs employed
Approve5.Anything else (borderline)
Refer to underwriterSenior Sign-off
Independent of the decision — you still pick the outcome, but flag the large loans so a senior underwriter co-signs.
Loan amount above $250,000
Senior sign-offStep-by-step Instructions
1. Open the test system and read the first application's full profile
2. Check default record, credit score, DTI, and employment history
3. Apply the policy in order — the first match is your decision
4. Select the reason that drove the decision
5. Flag the loan for senior sign-off if the amount exceeds $250,000
6. Submit and work through all 6 applications for your score
What you will learn
✓ Multi-factor risk reasoning with priority ordering
✓ Mapping numeric thresholds to confident decisions
✓ The Refer path — deferring borderline cases instead of guessing
✓ Human sign-off triggers for high-value exposure
✓ Audit-ready justification for every lending decision
✓ Real tool exposure: UiPath Document Understanding + Agents, Power Automate AI Builder, Automation Anywhere AI Agent Studio, Blue Prism decision stages
Common mistakes
• Approving an applicant who has a prior default
• Auto-approving a borderline profile instead of referring it
• Ignoring a high debt-to-income ratio because the score looks fine
• Forgetting senior sign-off on a large loan
• Declining a strong profile over a single soft factor
Complexity
Advanced · Agentic
Numeric risk reasoning, deferral logic, and human sign-off triggers.