When you submit a credit card or loan application, lenders may use automated models, traditional credit scoring, application information, credit-report data, or combinations of different underwriting methods to assess the application.
Understanding how automated credit decisioning works won't just satisfy curiosity — it can help you understand why you were approved or denied, and what you can do about it.
The Evolution from Scores to Models
For decades, lenders have commonly used FICO scores as one input in credit decisions. A FICO score is a three-digit number derived from information in a consumer's credit report. Lenders have historically used credit scores alongside other underwriting criteria, and lending standards vary by lender and product.
A FICO score captures only what's in your credit report. It doesn't know your income, your cash flow patterns, or other factors that may indicate ability to repay.
Modern automated credit decisioning may use credit scores as one input among many. Depending on the lender, product, applicable law, consumer permissions, and underwriting model, information considered may include:
- •Credit-report information (payment history, utilization, account age)
- •Application information
- •Income or employment information
- •Cash-flow or banking information when appropriately obtained
- •Trended credit information — not just your utilization today, but whether it's been rising or falling
- •Rental-payment information where available and applicable
- •Other permitted information used by the lender
How Machine Learning Models Work
Traditional credit models were rule-based: "if score > 700 and income > $40,000, approve." Machine learning models are different — they learn patterns from historical data.
Models may be trained on historical data to identify statistical relationships associated with repayment or credit risk. These patterns can be complex and non-linear — a relationship between two variables that no human analyst would have identified.
Different institutions may use different statistical or machine-learning techniques. The specific model and data vary by lender.
What This Means for Applicants
More nuanced approvals
Alternative data or more complex models may allow some lenders to evaluate applicants differently, including consumers with limited traditional credit histories. This can be particularly relevant for thin-file applicants, recent immigrants, and young adults.
Varying criteria among lenders
Underwriting criteria vary among lenders, and consumers generally cannot know every factor or weight used in a lender's proprietary model. Traditional credit guidance — such as keeping utilization low and making payments on time — remains broadly useful, but it may not capture everything a particular lender considers.
Less transparency
General categories influencing credit scores may be publicly described, while exact proprietary scoring models and lender underwriting systems are not necessarily public. Machine learning models can be complex and difficult to interpret. While regulations require lenders to provide adverse action reasons when denying applications, these reasons may not fully capture what a complex model actually weighted.
In this context, transparency means that a consumer can receive understandable and specific reasons for an adverse credit decision even when complex algorithms or AI are used. The CFPB has issued guidance clarifying that adverse action notification requirements apply equally to credit decisions involving complex algorithms, so consumers are entitled to know the principal reasons for the decision regardless of the technology behind it.
Potential for bias
Machine learning models learn from historical data, and historical lending has included discriminatory patterns. Models trained on biased historical outcomes can create fair-lending, data-quality, transparency, and discrimination concerns. These issues can arise even without explicitly using protected characteristics. Regulators are increasingly focused on this area.
ECOA, Regulation B, FCRA and AI
Under the Equal Credit Opportunity Act (ECOA) and Regulation B, creditors taking adverse action generally must provide the applicant with the specific principal reasons for the action, or provide notice of the right to obtain those reasons where permitted. This requirement still applies when complex algorithms or artificial intelligence are used. When adverse action is based on information from a consumer report, the Fair Credit Reporting Act (FCRA) imposes additional notice requirements.
The CFPB has signaled that it expects lenders using automated models to comply with the same fairness and transparency requirements as traditional scoring systems.
Practical Implications
If you're declined for credit by an automated system:
- •Review your adverse action notice — If your application is denied or another adverse action is taken, applicable federal law generally requires the creditor to provide reasons for the decision or information about your right to obtain them. For a detailed walkthrough of what to do after a denial, see our guide on what to do when your credit card application is denied.
- •Check your credit report — Errors in your credit file can affect credit decisions.
- •**If you authorized a lender to consider banking or cash-flow information, review that information for accuracy as well.**
- •Apply to different lenders — Different lenders use different models. A decline from one doesn't mean a decline from all. Keep in mind that applications may result in hard inquiries depending on the product and circumstances.
Automated and machine-learning models may create opportunities for more nuanced credit evaluation, but they also raise questions about accuracy, transparency, privacy and fair lending. Consumer-protection requirements continue to apply regardless of the technology used.
Sources & References
The following resources from official agencies provide authoritative information on adverse action requirements, AI in credit decisions, and consumer protections.
- •Consumer Financial Protection Bureau — CFPB Issues Guidance on Credit Denials by Lenders Using Artificial Intelligence
- •Consumer Financial Protection Bureau — Consumer Financial Protection Circular 2022-03: Adverse action notification requirements in connection with credit decisions based on complex algorithms
- •Consumer Financial Protection Bureau — Regulation B — § 1002.9 Notifications
- •Federal Trade Commission — Using Consumer Reports for Credit Decisions: What to Know About Adverse Action and Risk-Based Pricing Notices