DDPs in lending: A decision-making case study

Feb 23, 2025

Introduction

The lending industry presents a classic example of a Deep Domain Problem (DDP), where expert knowledge, regulatory requirements, and complex decision-making intersect. As outlined in How to Identify Deep Domain Problems Worth Solving, let's unpack how professionals navigate this domain by interviewing experts, recording workflows and gut calls, pinpointing inputs, identifying critical forks, and comparing approaches.

Step 1: Interview Experts and Record Workflows

Expert 1 (Underwriter): Focuses on risk mitigation

  • Steps: Check credit score, review financial statements, assess debt-to-income (DTI) ratio, verify collateral
  • Gut call: "If the business seems unstable despite good numbers, I dig deeper into cash flow"

Expert 2 (Loan Officer): Prioritizes client relationship and feasibility

  • Steps: Evaluate business plan, discuss repayment capacity with applicant, check industry trends
  • Gut call: "A passionate owner with a solid plan can outweigh a shaky credit history"

Expert 3 (Analyst): Relies on data models

  • Steps: Run credit risk algorithm, compare application to historical defaults, assess market conditions
  • Gut call: "Numbers don't lie, but outliers need context—like a sudden market dip"

Step 2: Pinpoint Inputs

As described in How to Break Down a Deep Domain Problem, we categorize key inputs:

Data Inputs

  • Credit score
  • DTI ratio
  • Revenue
  • Cash flow
  • Collateral value

Qualitative Inputs

  • Business plan quality
  • Owner credibility
  • Industry outlook

External Inputs

  • Market conditions
  • Historical loan performance

Step 3: Highlight Forks ("It Depends")

  • Credit score < 600: Immediate rejection vs. consider compensating factors?
  • High DTI but strong cash flow: Approve with conditions vs. reject?
  • Weak collateral but growing industry: Risk it vs. demand more security?

Step 4: Compare Approaches

Pattern

All experts check credit and financials, but weight varies:

  • Underwriter: 70% data
  • Loan Officer: 50% story
  • Analyst: 90% model

Gap

  • Underwriter skips industry trends
  • Loan Officer undervalues collateral
  • Analyst misses owner intent

Step 5: Decision-Making Tree

Following the principles from Ubiquitous Language: The Foundation for Solving Deep Domain Problems, here's a comprehensive decision tree:

START: Small Business Loan Application Received
├── Input 1: Credit Score
│   ├── ≥ 600
│   │   ├── Input 2: Debt-to-Income (DTI) Ratio
│   │   │   ├── ≤ 40%
│   │   │   │   ├── Input 3: Cash Flow (Positive last 12 months?)
│   │   │   │   │   ├── Yes
│   │   │   │   │   │   ├── Input 4: Collateral Value ≥ 50% of Loan?
│   │   │   │   │   │   │   ├── Yes → APPROVE LOAN
│   │   │   │   │   │   │   └── No
│   │   │   │   │   │   │       ├── Input 5: Industry Growth?
│   │   │   │   │   │   │       │   ├── Yes → APPROVE WITH HIGHER RATE
│   │   │   │   │   │   │       │   └── No → REJECT (Weak security)
│   │   │   │   │   └── No
│   │   │   │   │       ├── Input 6: Business Plan Quality
│   │   │   │   │       │   ├── Strong → APPROVE WITH MONITORING
│   │   │   │   │       │   └── Weak → REJECT (Unstable cash flow)
│   │   │   └── > 40%
│   │   │       ├── Input 7: Cash Flow Trend (Improving?)
│   │   │       │   ├── Yes
│   │   │       │   │   ├── Input 8: Owner Credibility
│   │   │       │   │   │   ├── High → APPROVE WITH CONDITIONS
│   │   │       │   │   │   └── Low → REJECT (Risky profile)
│   │   │       │   └── No → REJECT (High DTI, no recovery)
│   │   └── Input 9: Compensating Factors (e.g., Co-signer, Large Deposit)?
│   │       ├── Yes
│   │       │   ├── Return to Input 2 (Reassess DTI with adjustments)
│   │       │   └── If still weak → REJECT
│   │       └── No → REJECT (Credit too low)
│   └── < 600
│       ├── Input 10: Exceptional Circumstances (e.g., Recent bankruptcy recovery)?
│       │   ├── Yes
│       │   │   ├── Input 11: Strong Business Plan + Industry Boom?
│       │   │   │   ├── Yes → APPROVE WITH STRICT TERMS
│       │   │   │   └── No → REJECT
│       │   └── No → REJECT (Credit disqualifies)
└── END

Explanation of the Tree

Sequential Logic

Starts with a hard filter (credit score), reflecting the underwriter's risk focus, but allows exceptions (loan officer's influence).

Critical Forks

"It depends" moments like high DTI with improving cash flow or weak collateral in a hot industry capture expert trade-offs.

Tacit Smarts

Gut calls surface—e.g., "owner credibility" (Loan Officer) tempers data-driven rejection (Analyst), while "industry growth" (Analyst) offsets weak collateral (Underwriter).

Patterns and Gaps

The tree balances data (credit, DTI) with narrative (business plan, owner), bridging the experts' approaches while exposing gaps (e.g., no one prioritizes market downturns explicitly).

Conclusion

This decision tree is a practical map of how lending pros navigate approvals, blending hard rules with contextual judgment—exemplifying the principles discussed in The Evolving Role of Programmers in Solving Deep Domain Problems. For more insights on breaking down similar complex domain challenges, see our Case Study: Decomposing a DDP in Supply Chain Optimization.