Automation, Artificial Intelligence, and Machine Learning in Consumer Lending
May 10, 2023
Automation, Artificial Intelligence, and Machine Learning in
Consumer Lending
Financial firms may use algorithms—pre-coded sets of
fintech products, making it a subject of increased interest
instructions and calculations that are executed
for the public and policymakers.
automatically—to enhance consumer loan underwriting, the
process of evaluating the likelihood that applicants will
ML Models in Consumer Loan Underwriting
make timely loan repayments. Lenders may rely upon
Consumer loan underwriting can potentially be enhanced
forms of automated analysis to help decide whether to offer
by ML models. ML models could improve efficiency and
consumers loans and at what terms. Faster computing
performance and reduce costs for financial institutions,
power, internet-based products, and cheaper data storage at
potentially expanding credit access or making credit less
scale have increased the prevalence of algorithms.
expensive for some consumers. ML models could make
consumer underwriting decisions more accurate by
This In Focus discusses developments in automated
identifying new patterns, such as changing credit
decisionmaking, artificial intelligence (AI), and machine
conditions, and by automatically updating the models to
learning (ML) in consumer loan underwriting. First, it
make more accurate underwriting assessments.
focuses on market developments, then it discusses the
current regulatory framework, then finally, it highlights
However, ML models can also introduce risks. One risk is a
selected policy issues.
lack of explainability, the inability to explain why programs
make particular decisions. Another risk is dynamic
Market Developments
updating, which is when models evolve over time without
Since the 1970s, consumer loan underwriting has become
oversight. ML models also raise concerns that they may not
more automated, first with the increasing use of credit
perform as intended, possibly resulting in higher loan losses
scores and more recently with new data and technologies.
in new market environments or discrimination against
Credit scores are a (numeric) metric calculated with
protected groups.
information in consumer credit reports and prepared for
lenders to determine the likelihood of loan default. New
Current Federal Regulatory Framework
technological innovations have been used to update
The Consumer Financial Protection Bureau (CFPB) is the
automated processes, in some cases beyond traditional
primary consumer protection regulator for consumer
numeric credit scores. For example, for some lenders, the
financial products and services. One of the CFPB’s
internet has been incorporated to accept applications, and
statutory objectives is to ensure that “markets for consumer
new data sources are used to conduct consumer loan
financial products and services operate transparently and
underwriting. Alternative data generally refers to
efficiently to facilitate access and innovation.” The CFPB
information that may be used to determine a consumer’s
has the authority in consumer financial markets to write
creditworthiness that the national credit reporting
regulations and enforce the law for both bank and nonbank
agencies—Equifax, Experian, and TransUnion—have not
financial institutions. However, the CFPB’s supervisory
traditionally used when calculating credit scores for
authority to examine financial institutions for consumer
consumers. Further, AI and ML technologies have
protection compliance varies based on the charters,
advanced rapidly in recent decades. AI technologies are
activities, and size of institutions. Therefore, financial
computerized systems that work and react in ways
regulators may monitor some nonbank fintech companies
commonly thought to require intelligence, such as solving
less than traditional banks.
complex problems in real-world situations. ML is often
referred to as a subfield of AI with algorithms designed to
Regulatory Uncertainty
automatically improve their performance through
Many financial laws and regulations that existed prior to
experience with little or no human input.
recent ML technological developments have led to
questions concerning their effectiveness achieving their
These technological developments potentially allow for
designed policy goals as these potentially beneficial
greater speed, accuracy, and confidence in loan decisions.
technologies evolve. Relevant laws and regulations may
They are currently used more frequently in fintech products
need to be reconsidered or updated in response to the future
than in more traditional consumer lending products,
use of ML models in consumer loan underwriting. This
particularly ML models and alternative data. Fintech (short
often involves balancing efforts to encourage innovation
for financial technology) refers to advances in technology
while protecting consumers.
incorporated into financial products and services. Many
companies—both traditional financial firms and new
Federal financial regulators have been monitoring ML
technology-focused entrants to the market—are developing
models in consumer lending. In March 2021, the bank and
credit union regulators, along with the CFPB, requested
https://crsreports.congress.gov
Automation, Artificial Intelligence, and Machine Learning in Consumer Lending
information on financial institutions’ use of AI, including
Underserved Consumers and Access to Credit
ML models. In April 2023, the CFPB, along with other
In the United States, robust consumer credit markets allow
federal agencies, published a joint statement emphasizing
most consumers to access financial services and credit
that existing legal authorities still apply to automated
products to meet their needs in traditional financial markets.
systems and AI.
However, consumers who have difficulty entering the
traditional credit reporting system face challenges accessing
Consequently, some financial institutions, particularly
many consumer credit products, because lenders are unable
many banks or other highly regulated parts of the financial
to assess their creditworthiness. Limited credit history is
system, may choose not to use ML models to approve or
correlated with age, income, race, and ethnicity, and many
reject applicants for consumer credit, even if they are more
of these consumers are young.
accurate or efficient, due to regulatory uncertainty and
compliance risks.
Automated underwriting and ML models may expand credit
access or make credit less expensive for some consumers.
Policy Issues
In particular, these technologies may increase financial
The use of ML algorithms in credit underwriting has raised
inclusion for younger consumers, who may be more likely
a number of policy issues of interest for financial regulators
to have limited credit histories in the credit reporting
and Congress.
system.
ML Models and Explainability Concerns
Data Privacy, Security, and Transparency
The ability of regulators or other outside parties to
In credit underwriting, ML models often access sensitive
understand what an ML program did, and why, may be
consumer financial data, and the increase in digital data
limited or nonexistent. This poses a significant challenge
collection raises greater privacy and cybersecurity
for companies using ML programs to ensure that they will
concerns. These data practices raise questions over what
produce outcomes that comply with applicable laws and
consumer information is appropriate to collect and use for
regulations.
loan underwriting.
When a lender denies a loan application, the lender must
Laws such as the Fair Credit Reporting Act (FCRA; 15
send an adverse action notice to the applicant explaining the
U.S.C. §1681) and the Gramm-Leach-Bliley Act (GLBA;
reason for the denial. Some question how well lenders will
P.L. 106-102) impose requirements on firms that use
understand and be able to explain the reasons for adverse
consumer data for credit underwriting. As data use in
actions resulting from ML algorithms. To address this issue,
consumer financial services has grown, some have debated
some observers assert that regulators should set standards
whether the scope of these laws should be expanded.
for how ML programs are developed, tested, and
monitored, although debate exists about what these
CRS Resources
standards should include. Concerns exist about ML model
CRS Report R47475, Consumer Finance and Financial
fairness; the ability to provide more algorithm transparency;
Technology (Fintech), coordinated by Cheryl R. Cooper.
and developing processes to assess ML models, for
example, for fairness, reliability, privacy, and security. In
CRS Report R46795, Artificial Intelligence: Background,
May 2022, the CFPB issued guidance clarifying that
Selected Issues, and Policy Considerations, by Laurie A.
lenders using “complex algorithms” still need to comply
Harris.
with adverse action notice requirements.
CRS In Focus IF11630, Alternative Data in Financial
Algorithmic Bias and Fair Lending
Services, by Cheryl R. Cooper.
Consumer loan underwriting models using ML can
introduce fair lending risks due to biases in data or model
CRS Report R44125, Consumer Credit Reporting, Credit
development. ML models may have training data biases,
Bureaus, Credit Scoring, and Related Policy Issues, by
which is when a model has biases due to the limited or
Cheryl R. Cooper and Darryl E. Getter.
flawed dataset it was developed on. Historical data can
reflect historical biases, potentially creating models that
CRS In Focus IF10031, Introduction to Financial Services:
discriminate against protected classes. In addition,
The Consumer Financial Protection Bureau (CFPB), by
alternative data may include proxies for protected classes
Cheryl R. Cooper and David H. Carpenter.
that might also bias ML models.
CRS In Focus IF11195, Financial Innovation: Reducing
The Equal Credit Opportunity Act (ECOA; 15 U.S.C.
Fintech Regulatory Uncertainty, by David W. Perkins,
§§1691-1691f) generally prohibits discrimination in credit
Cheryl R. Cooper, and Eva Su.
transactions based upon certain protected classes, including
sex, race, color, national origin, religion, marital status, age,
Cheryl R. Cooper, Analyst in Financial Economics
and “because all or part of the applicant’s income derives
from any public assistance program.” Questions exist about
IF12399
how lenders can comply with ECOA and other fair lending
laws when using ML models for loan underwriting.
https://crsreports.congress.gov
Automation, Artificial Intelligence, and Machine Learning in Consumer Lending
Disclaimer
This document was prepared by the Congressional Research Service (CRS). CRS serves as nonpartisan shared staff to
congressional committees and Members of Congress. It operates solely at the behest of and under the direction of Congress.
Information in a CRS Report should not be relied upon for purposes other than public understanding of information that has
been provided by CRS to Members of Congress in connection with CRS’s institutional role. CRS Reports, as a work of the
United States Government, are not subject to copyright protection in the United States. Any CRS Report may be
reproduced and distributed in its entirety without permission from CRS. However, as a CRS Report may include
copyrighted images or material from a third party, you may need to obtain the permission of the copyright holder if you
wish to copy or otherwise use copyrighted material.
https://crsreports.congress.gov | IF12399 · VERSION 1 · NEW