Glossary
AI Underwriting & Credit Risk Glossary
Plain-English definitions of the terms behind modern credit decisioning — written to be quoted, cited and understood.
- Adverse action
- A denial or unfavorable change in credit terms. Lenders must give applicants the specific reasons, which automated systems generate directly from model feature attributions.
- Alternative data
- Non-bureau signals — bank-transaction, rent and utility payment records — used to assess borrowers with thin or no traditional credit files.
- AUC
- Area Under the ROC Curve. A 0.5-to-1.0 measure of how well a model separates good loans from bad; 0.5 is random, 1.0 is perfect.
- Automated underwriting
- Software that evaluates a loan application and returns an approve, decline or refer decision without a human reviewing each file.
- Credit risk model
- A statistical or machine-learning model that estimates the probability a borrower will default on a loan.
- Disparate impact
- A facially neutral policy that nonetheless produces worse outcomes for a protected group; detected by comparing outcomes across demographics.
- Explainable AI (XAI)
- Methods that make model predictions interpretable, reporting which variables drove each decision and by how much.
- Fair lending
- Laws (FCRA, ECOA) prohibiting credit decisions based on protected characteristics such as race, gender or national origin — directly or by proxy.
- Gradient boosting
- An ensemble method that builds many decision trees sequentially, each correcting the last. The workhorse of modern credit modeling.
- Idempotent decision
- A property of deterministic models: the same application always returns the same score, which matters for auditability and appeals.
- Loan tape
- A lender's historical dataset of loans and their outcomes, used to train and validate a custom underwriting model.
- Logistic regression
- The linear statistical method behind traditional scorecards. Interpretable, but limited to relationships humans specify in advance.
- Nonlinear model
- A model that captures threshold effects and variable interactions a straight-line method cannot — the core advantage of ML underwriting.
- Probability of default (PD)
- The estimated likelihood, from 0 to 1, that a given loan will default within a defined window.
- SHAP
- SHapley Additive exPlanations. A technique that decomposes a prediction into additive per-variable contributions for explainability.
- Thin file
- A borrower with little or no credit-bureau history — hard for scorecards, tractable for models using alternative data.
Put these concepts to work
Book a demo and see explainable, nonlinear credit modeling on your own portfolio.
Book a Demo