More loan approvals at lower rates in a simulation

AI system

System: Upstart Network underwriting and pricing model
Model: Not assessed
Release: Not assessed
Application developer: Upstart Network, Inc.
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Alternative data combined with machine-learning underwriting, compared against a hypothetical traditional model on Upstart's own applicant pool. A regulatory compliance plan requiring fair-lending testing and reporting.
Human oversight: Not assessed
Relation to AI: The CFPB states that the reported gains reflect the net effect of both the alternative data and the machine-learning methodology. So the model's contribution cannot be separated from its data. Upstart developed and deployed the model and supplied the comparison. The independent monitor later found approval disparities for Black applicants and disagreed with Upstart about a less discriminatory alternative.
AI contribution: The CFPB states that the reported gains reflect the net effect of both the alternative data and the machine-learning methodology. So the model's contribution cannot be separated from its data. Upstart developed and deployed the model and supplied the comparison. The independent monitor later found approval disparities for Black applicants and disagreed with Upstart about a less discriminatory alternative.

Outcome identifiers: Not assessed
Benefit: approving 27% more applicants at 16% lower APRs than a traditional model
Category: Increased Inclusion
Status: Not assessed
Beneficiary groups: People. Society.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: approving 27% more applicants at 16% lower APRs than a traditional model
Comparison: a hypothetical model using traditional application and credit file variables without machine learning
Results: Reference: approval rate and average APRs of the traditional model (not quantified). Result: approves 27% more applicants than the traditional model, and yields 16% lower average APRs for approved loans.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Participants: Developer: Upstart Network, Inc. Operator: Upstart Network, Inc.
Jurisdictions: Developer: United States. Operator: United States.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Conditions for replication: An applicant pool and comparison model similar to the 2019 test. Fair-lending testing that measures approval as well as pricing disparities. Independent monitoring, which in 2024 found approval disparities the 2019 test did not report.
Replication attempts: 2 follow-up entries. Outcome unresolved: 2.
Later findings: 27 March 2024. The independent monitor found approval disparities for Black applicants and proposed a less discriminatory alternative. Upstart and the monitor did not agree on adopting it. The earlier termination of the no-action letter was an administrative action, not litigation.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Source numbers: AIBD-0175-source-0001–0003
Citations: Numbered references below.
Language: Not assessed
Source relationships: Upstart supplied the approval and APR estimates. CFPB published them without replicating them. The later independent monitor examined fairness and identified approval disparities for Black applicants. [1] [2] [3]
Original evidence: Upstart supplied the approval and APR estimates. CFPB published them without replicating them. The later independent monitor examined fairness and identified approval disparities for Black applicants. [1] [2] [3]
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Assessed benefits: People. Society.
Depth: Not assessed
Reasons: People: Approval and pricing effects are reported as aggregate rates over Upstart's applicant pool, with no individually attributable outcome. Society: Sources describe aggregate access effects and later disparities; no societal outcome is measured.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Listed, not verified

Kind: Not assessed
Related entries: Not assessed
Field: Consumer credit. Unsecured personal loans and credit access.
Purpose: Underwrite and price unsecured consumer loans using alternative data and machine learning.
Dates: Date basis unassessed: 2019 (precision unassessed). Updated: 25 October 2026.
Stage: Not assessed
Version: SHA-256: 058110bb515d…