AI Trend Research

← AI glossary

AI glossary · Basic · Classic ML, recsys & search · Finance & insurance

Credit scoring model

A credit scoring model calculates from data how likely a person or business applying for a loan or card is to repay on time, and expresses it as a score or grade. Financial companies use this score to decide approval, credit limits and interest rates.

About this term
Level
Basic · Technical terms you often meet in AI news
Category
Industry AI
Added
2026-10-10
In plain words It is like putting into numbers the sense you get of whether someone is trustworthy by looking over their record of keeping past promises before lending money to a stranger. But unlike human intuition, a model can be badly wrong in situations that were not in the past data it learned from.

How it works

The model is trained on past borrowers' information (repayment history, income, debt level, length of relationship, etc.) paired with whether they actually went delinquent. It looks at a new applicant's information, predicts the probability of delinquency, and converts this into a score or grade. Traditionally, easy-to-interpret statistical models such as logistic regression were widely used; today machine learning techniques such as gradient boosting are also used. The final decision combines the score with the company's lending policies and rules.

How a credit scoring model produces a score

  1. 1
    Collect past dataRepayment history · income · debt and delinquency
  2. 2
    Train the modelRegression · gradient boosting, etc.
  3. 3
    Predict delinquency probabilityInput new applicant information
  4. 4
    Convert to score · grade
  5. 5
    Decide approval · limit · interest rateApplied together with company policy
A model trained on past data predicts a new applicant's probability of delinquency, and the decision is made together with company policy.

Why it matters

It is a key tool that lets financial companies reduce losses while still lending to people who are able to repay. Automating underwriting allows large numbers of applications to be processed quickly and consistently. Using alternative data such as phone bill and utility payment records also opens more ways for people with thin financial records, such as young people new to the workforce and the self-employed, to be assessed.

Good to know

A model can learn biases contained in past data as they are, so it must be checked for unfavorable outcomes for particular groups. Because lenders often have to explain to rejected applicants why they were turned down, explainability is especially important. Data drift, where changing economic conditions shift the data distribution and degrade model performance, is also common. That is why the financial sector regularly validates and monitors models within a model risk management framework.

Example

When you check your loan limit in an internet bank or fintech app, the limit and interest rate appear within seconds; behind the scenes, a credit scoring model calculates the likelihood of repayment from the applicant's information. Some services underwrite by combining scores from credit bureaus with models built on their own data.

Related terms

Alternative data Model risk management Gradient boosting Regression analysis Data drift Interpretability