Earnings Show AI Lending Platforms Scaling Originations

The latest earnings cycle from AI-native lending platforms reveals an industry inflection: machine learning underwriting has moved from experimental advantage to dominant competitive strategy. Upstart reported 178% origination growth, Pagaya posted 143% year-over-year increases, and Figure's AI-powered HELOC platform crossed $5 billion in total originations — all while maintaining or improving credit performance metrics.

For accounting and finance professionals advising lending clients or evaluating fintech credit exposure, these numbers signal that AI underwriting is no longer a niche innovation but a structural market shift that traditional lenders must match or face permanent market share erosion.

The Earnings Data

Upstart (Q4 2025 earnings):

• Origination volume: $4.8B (+178% YoY)
• Revenue: $219M (+142%)
• Default rate: 4.2% (versus 5.8% for FICO-equivalent cohort)
• Approval rate: 62% of applications (versus ~40% at traditional lenders for same demographics)
• Average decision time: 8 seconds

Pagaya (Q4 2025):

• Network volume: $2.9B (+143% YoY)
• Revenue: $287M (+98%)
• Partner institutions using Pagaya AI: 38 banks and credit unions (+52% YoY)
• Loan performance: 22% fewer delinquencies than partner institutions' own underwriting

Figure (Q4 2025):

• HELOC originations: $1.4B (+120% YoY)
• Total platform originations: $5.2B cumulative
• Average HELOC closing time: 5 days (versus 45 days industry average)
• Default rate: 0.8% (versus 1.2% industry average for HELOC)

The consistent pattern: AI lenders grow originations 120-180% annually while simultaneously improving credit quality — a combination that traditional underwriting simply cannot achieve because FICO-based decisioning treats credit quality and volume as trade-offs (approve more → worse performance).

Why AI Underwriting Wins

Traditional credit scoring (FICO, VantageScore) evaluates 20-30 variables derived primarily from credit bureau data: payment history, credit utilization, length of history, mix of credit, and new inquiries. This framework was designed in the 1980s and hasn't fundamentally changed.

AI underwriting evaluates 1,000-1,500+ variables including:

Banking behavior: Transaction patterns, savings consistency, income stability over time, overdraft frequency, and spending-to-income ratios derived from bank account access (via Plaid, MX, or open banking APIs).

Employment signals: Job tenure patterns, industry stability, employer financial health (public company data), income trajectory versus static current income.

Payment velocity: How quickly applicants pay non-credit obligations (rent, utilities, subscriptions) — a behavioral signal that's more predictive than whether they've previously defaulted.

Alternative data: Rent payments, utility payment history, subscription maintenance, educational credentials — expanding the assessment to populations that traditional scoring underserves.

Real-time market data: Local employment conditions, housing market trends, industry-specific risk factors that affect the applicant's ability to repay based on their specific circumstances.

The result: AI can identify creditworthy borrowers that FICO-based systems reject (expanding the market) while more precisely identifying risky borrowers that FICO-based systems approve (improving performance). This dual improvement explains how AI lenders grow volume AND reduce defaults simultaneously.

The Competitive Threat to Traditional Lenders

The earnings data quantifies the competitive gap:

Speed differential. 8 seconds (Upstart) versus 3-5 days (traditional underwriting). For point-of-sale lending, auto financing, and time-sensitive personal loans, speed determines which lender wins the customer.

Approval differential. 62% approval rate (Upstart) versus ~40% (traditional) for the same applicant pool. Every declined applicant at a traditional lender is a potential customer for AI platforms — and increasingly, they know it.

Cost differential. AI underwriting costs $3-8 per decision versus $50-200 for human-reviewed traditional underwriting (including labor, time, and infrastructure). This 10-25x cost advantage enables AI lenders to profitably serve smaller loan sizes that traditional lenders can't economically underwrite.

Performance differential. AI default rates are 15-25% lower on equivalent-risk cohorts. Over a loan portfolio's lifetime, this translates to hundreds of basis points of additional margin — earnings that can fund lower interest rates (winning customers) or higher profitability (winning investors).

Traditional banks are responding by licensing AI underwriting platforms (38 banks now use Pagaya) rather than building internally — implicitly conceding that their technology organizations can't replicate a decade of ML model development.

Why This Matters

For CFOs at lending institutions: If your institution still relies primarily on FICO-based decisioning, you're declining creditworthy applicants that AI lenders capture — and possibly approving risky applicants that AI would reject. Evaluate build-versus-buy options for AI underwriting before market share erosion becomes irreversible.

For accounting firms auditing loan portfolios: AI-underwritten portfolios require different audit approaches than FICO-based books. Model validation, backtesting documentation, fair lending compliance (disparate impact of ML models), and reserve adequacy calculations all need methodology updates for AI-driven credit decisions.

For regulators: The 35% approval rate gap between AI and traditional lenders raises access-to-credit policy questions. If AI lenders safely approve applicants that traditional lenders reject, the traditional system is arguably denying credit to qualified borrowers — a fair lending concern in reverse.

For investors in bank stocks: The 120-180% origination growth rates at AI lenders represent direct market share capture from traditional institutions. Banks that don't adopt AI underwriting within 2-3 years will show declining origination volumes that compress earnings per share.

For consumer finance advisors: Clients rejected by traditional lenders should be directed to AI-powered alternatives where they may receive approval based on holistic financial behavior rather than backward-looking credit scores.

The key takeaway: AI lending platforms growing originations 120-180% with 15-25% lower defaults prove that machine learning has permanently disrupted credit underwriting — traditional lenders that don't adopt AI within 2-3 years will lose the most creditworthy borrowers while retaining the riskiest ones, a portfolio death spiral.