Lenders often use credit scoring and credit underwriting as if they mean the same thing, and they do not.
Credit scoring puts a number or a risk band on an applicant. Credit underwriting is the full process of deciding whether to lend, how much and on what terms, and a score is only one input into it.
This article explains each term, the main differences between them, eight steps in which they work together, and how a lender can set up both well.
What credit scoring is
Credit scoring turns information about an applicant into a number or a risk category that predicts how likely they are to repay.
The CFPB describes a credit score as a number used to predict how likely someone is to pay back a loan on time, and says lenders use it to decide whether to offer credit, what interest rate to charge and what credit limit to set.
Scores come in different forms, from fixed scales such as 300 to 850 to internal bands such as low, medium and high risk, so no single scale applies to every lender.
A scoring model is built from past repayment data. According to Wikipedia’s overview of consumer credit risk, a scorecard attributes a number to a customer that indicates the predicted probability of a certain behavior, using data from the application, credit reference agencies or products the customer already holds with the lender.
A lender might find that customers who repay earlier loans on time, carry manageable debt and show regular cash inflows default less often, and the model uses that pattern to estimate how a new applicant might behave.
A score has limits, because it reflects the data and assumptions behind it. A borrower with little recorded credit history can score low simply because the model has little to go on.
The World Bank’s Credit Reporting Knowledge Guide describes credit bureaus adding data such as utility and telephone payments so that people without formal credit can build a history, which shows how much the quality of the data matters.
What credit underwriting is
Credit underwriting is how a lender decides whether to lend, how much, and on what terms. The OCC, a US bank regulator, lists the five Cs of credit in its handbook on installment lending: character, capacity, capital, collateral and conditions.
Capacity is the borrower’s ability to repay. In its handbook on residential real estate lending, the OCC says a mortgage underwriter’s job is to approve or deny the application, and that underwriting standards typically include a credit bureau score or an application score. So a score is one input, and the underwriter makes the decision.
Underwriting can be done by people, by software or by both. At a traditional lender, a credit officer might read bank statements, financial records, collateral documents and a bureau report before recommending a decision.
At a digital lender, software might check identity, pull credit data, work out affordability and approve eligible applications automatically.
Here is an example. A trader applies for a 2,000,000 loan, and her credit score puts her in a medium-risk band. When the lender looks at her bank account, it sees that most of her recent deposits came from a single transfer, not from regular sales.
Her balance looks healthy, but her shop may not earn enough to cover the monthly repayment. The underwriter asks where the money came from, looks at her usual monthly sales and offers 1,000,000 over a longer period instead.
The main differences
Scoring measures risk, and underwriting makes the decision. A score tells the lender how likely an applicant is to repay, based on a few characteristics of the borrower and the loan.
Underwriting looks at the whole application and ends with an answer: approve, decline, ask for more information or approve with conditions. For example, a score might label an applicant medium risk, and the underwriter then decides whether to lend 1,000,000, lend less or ask for more documents.
The two also use different information. A scoring model works from structured data such as repayment history, debts and account activity.
Underwriting can use that data and more, including documents, financial statements, collateral and answers to follow-up questions.
A person, a decision engine or both can do the underwriting. For example, a model can score a salaried borrower from her repayment history alone, but an underwriter might also ask for payslips and an employer letter before approving a larger loan.
Unusual cases show the difference most clearly. A scoring model applies what it learned from past data and cannot ask questions. Underwriting can investigate. Say a bureau report shows a loan the applicant does not recognize.
An underwriter can find out whether the loan is hers, whether the report is accurate and how the lender’s policy treats it.
Lenders should keep a record of how scores influence decisions and when an application needs a closer look, so they can explain later why a loan was approved or declined.
Read more: How AI credit scoring works in lending (and where it fails)
How scoring and underwriting work together
Step 1: Receive the application
The lender collects information about the applicant, the amount requested, the purpose and the repayment period. For an individual, that means identity details, employment, income and existing obligations.
For a small business, it adds registration details, trading history, turnover and bank statements. Missing or inconsistent details should lead to a request for clarification and not an automatic rejection.
Step 2: Verify identity and screen for fraud
Before relying on the data, the lender confirms the applicant is who they claim to be. In Nigeria, for example, lenders may use authorized verification services involving the Bank Verification Number (BVN) or National Identification Number (NIN).
Checks can also look for the same identity details across several applications, altered documents and mismatched account ownership. Fraud screening and scoring are separate jobs.
An applicant with an excellent repayment history can still submit a fraudulent application, and a genuine applicant with a thin history needs further assessment and not an automatic rejection.
Step 3: Gather credit and financial information
The lender then collects what it needs to assess repayment risk, such as credit bureau reports, existing loan balances, repayment history, account transactions and declared income.
Where it is lawful and the customer has consented, lenders may add alternative data, such as transaction histories from digital financial services. Bureau data has limits, because records can be incomplete or out of date.
Step 4: Calculate the credit score
The scoring model looks at selected details, such as missed payments, current debts, how long the borrower has had credit and recent account activity, and it produces a score or risk band.
Suppose an applicant lands in the medium-risk band. That tells the lender how closely to look at the application under its policy, and it does not yet say whether the amount she wants is affordable.
A risk score and a probability of default are also related but different. A score ranks applicants by relative risk, while a probability of default estimates how likely a borrower is to default over a set period under a particular model.
Step 5: Assess affordability and repayment capacity
Now the lender checks whether the borrower can afford the repayments. It looks at income, essential expenses, existing debt payments, cash flow and the size and timing of the installments.
A trader who averages 900,000 in monthly sales may still struggle with a 200,000 monthly repayment once she pays for stock, rent, staff and household costs.
A farmer or seasonal business may earn most of its income at harvest, so a monthly schedule can clash with when the money arrives, and the loan structure should match the cash-flow cycle.
Step 6: Apply lending policy and investigate exceptions
The lender now compares the score with its underwriting policy, which may set minimum eligibility rules, limits on exposure and debt, documentation standards and the cases that need manual review.
A borrower with a good score and several loans already may exceed the lender’s affordability limit, so the lender might decline or offer a smaller amount.
A borrower who has never used formal credit may score weakly, and if policy allows, the lender can review verified transaction records or business cash flow to see whether she qualifies.
The lender should write down the reason for each exception and make sure staff cannot skip approval requirements without the right authority.
Step 7: Decide and set the loan terms
After the checks, the lender approves, declines or asks for more information. An approval comes with an amount, rate, repayment period, fees and conditions that fit the product, the borrower’s capacity, the rules and the lender’s view of risk.
The lender may approve less than the borrower asked for if the full amount would be too heavy, and it should explain the decision and terms clearly. It should also keep a record of the information it used and who or what made the decision.
The CFPB’s Circular 2022-03 says that in the United States, creditors that use complex algorithms must still give the specific principal reasons when they decline an applicant.
Step 8: Monitor performance and review the model
The process continues after the loan is paid out. The lender tracks repayments, missed installments and early signs of trouble, and uses the results to check that its scoring model and underwriting rules are working.
If borrowers in one risk band default more often than expected, the lender should look for the cause, which might be poor data, an economic shock, weak affordability checks, fraud or a model that no longer fits how borrowers behave.
Repayment data can also show that the lender has been declining people who would have repaid.
How Lendsqr supports scoring and underwriting
On Lendsqr, a lender builds a decision model, which is a set of rules that reflect its own risk criteria.
The model can use data from one or several credit bureaus, ecosystem data on borrowers across Lendsqr, and bank data shared by the borrower through open APIs, as the decision model page explains.
If a borrower fails a check under the lender’s credit risk rule, the system can decline the request automatically, and lenders can choose to send some applications to manual review instead, as Lendsqr’s auto-decline guide describes.
Fintechs with their own apps can add identity checks, credit decisioning and direct debit through the developer APIs. To try the platform, sign up for a Lendsqr account.
Read more: Frequently asked questions about alternative credit scoring
What to remember about scoring and underwriting
A credit score tells a lender how likely an applicant is to repay, and underwriting uses that estimate together with affordability, documents, policy and judgment to decide whether to lend and on what terms.
The main points to remember are to treat a score as one input, check identity and fraud separately, test affordability with real cash-flow evidence and send unusual cases to a person.
Lenders that then compare their model and rules with actual repayment will keep improving both, and each decision they make will be easier to explain.
