Affirm pivots to AI-driven underwriting to unlock growth through advanced pattern recognition in consumer credit behavior

Affirm Holdings, the San Francisco-based fintech pioneer, has officially deployed a sophisticated transformer-based underwriting model across its United States checkout ecosystem, marking a significant evolution in how the buy-now-pay-later (BNPL) provider assesses creditworthiness. By moving away from static, traditional credit scoring methods and toward a dynamic, time-sensitive analytical framework, Affirm aims to capture a more nuanced understanding of consumer financial health. This technological transition arrives at a critical juncture for the firm as it seeks to expand its reach deeper into its existing customer base while maintaining stringent risk management protocols.

The implementation of this new model, which occurred last week, represents a fundamental shift in the company’s underlying data strategy. Rather than relying solely on the snapshot provided by a FICO-style credit score, the transformer-based model analyzes the sequence and timing of credit events. This granular approach allows Affirm to differentiate between consumers who may share similar current debt burdens but are moving in opposite financial directions. Early performance data released by the firm indicates that this capability has already enabled the approval of applicants who would have been rejected under the previous machine-learning system, resulting in a 3.4% increase in completed purchases compared to a control group of similar risk profiles.

The Evolution of Affirm’s Lending Architecture

Since its inception, Affirm has sought to disrupt traditional retail credit by offering transparent, installment-based lending at the point of sale. However, the company has spent the better part of the last decade building an integrated financial ecosystem that extends well beyond simple retail financing. This roadmap includes the introduction of the Affirm Card, a digital marketplace within its proprietary app, and the Affirm Money Account—a high-yield savings product.

This vertical integration has transformed Affirm from a single-product lender into a comprehensive consumer finance platform. As consumers utilize the platform for more frequent and smaller-ticket transactions, the velocity of data flowing through Affirm’s servers has increased exponentially. In fiscal year 2026, the company reported 27.8 million active consumers, representing a 21% year-over-year growth. More importantly, the frequency of use has surged, with transactions per active consumer rising from 5.8 to 7.0. As the platform scales, the quality and accuracy of the underwriting engine become the primary determinants of the company’s long-term profitability and sustainable growth.

Moving Beyond the Static Score

The traditional credit scoring industry has long relied on the compression of complex financial histories into a three-digit number. While this method offers ease of use for legacy banking systems, it inherently loses the "temporal context" of a borrower’s life. A consumer who experienced a single missed payment two years ago but has since stabilized their finances may look identical on paper to someone who has missed payments in three consecutive months.

Affirm’s move to leverage transformer models—a class of artificial intelligence architecture originally designed for natural language processing—allows the company to treat a consumer’s financial history as a sequence of events. Just as a transformer model can understand the relationship between words in a sentence, Affirm’s new underwriting engine understands the relationship between events in a credit timeline.

"The time dimension of multiple purchases, multiple credit events in a customer’s life wasn’t being represented with particularly high fidelity," Affirm President Libor Michalek noted in a recent discussion regarding the firm’s technical strategy. By focusing on the trajectory of a consumer’s financial habits rather than just their current state, Affirm is effectively identifying "recoverers"—individuals whose creditworthiness is improving—before traditional systems catch up.

Chronology of Technological Integration

The transition to this model did not occur in a vacuum. It is the culmination of several years of internal research and development aimed at refining the firm’s machine-learning capabilities.

• 2021-2022: Affirm solidified its core BNPL offering, establishing strong partnerships with major retailers like Amazon and Shopify. During this period, the firm began accumulating the vast, high-frequency data sets necessary to train more advanced models.
• 2023: The company shifted focus toward its "super-app" strategy, launching the Affirm Card and expanding its savings products. This created a broader data loop, capturing not just how consumers spend, but how they save and manage cash flow.
• 2024: Affirm initiated rigorous back-testing of transformer-based models against its legacy machine-learning architecture. The focus was on identifying "false negatives"—qualified borrowers who were being excluded due to model rigidity.
• Late 2025/Early 2026: The firm moved to full production deployment, integrating the transformer model into its real-time checkout underwriting flow.
• Current Status: The company is now monitoring the performance of the model in live environments, specifically looking at default rates and consumer repayment behavior relative to the 3.4% increase in approved transactions.

Market Implications and Risk Management

For investors and analysts, the deployment of this model serves as a proof-of-concept for the "data moat" that Affirm has built. In the competitive BNPL landscape, where margins are often thin, the ability to approve more customers without increasing the overall risk profile of the loan portfolio is the ultimate competitive advantage.

However, the shift also invites scrutiny. Critics of AI-driven underwriting often point to the "black box" problem, where the decision-making process of a model becomes so complex that it is difficult for regulators to audit for bias or fairness. Affirm maintains that its deployment includes robust guardrails and that the model is designed to comply with existing fair-lending regulations. By identifying patterns that conventional models miss, Affirm argues it is actually increasing financial inclusion for individuals who are "credit invisible" or unfairly penalized by antiquated scoring systems.

Furthermore, the integration of this technology has implications for the broader retail sector. As Affirm’s underwriting engine becomes more precise, retailers partnering with the firm may see higher conversion rates at the checkout, particularly among segments that were previously underserved. If the 3.4% increase in purchases proves sustainable across different economic cycles, it could provide a significant tailwind for Affirm’s revenue growth in the coming quarters.

Official Perspective and Future Outlook

While Affirm has not publicly disclosed the exact computational parameters of its new model, the company’s leadership has been vocal about the necessity of this upgrade. The focus remains on "fidelity"—the ability to see the granular detail of a consumer’s financial life.

Industry analysts suggest that this pivot is a response to the changing macroeconomic environment. With interest rates remaining a focal point for consumer spending, lenders are under pressure to be more selective while simultaneously needing to grow their loan books. The transformer model offers a way to thread this needle: being more selective about who is approved, while being more generous to those who show positive, stable credit behavior over time.

As Affirm moves into the next phase of its fiscal year, the success of this model will be measured by its impact on the firm’s "provision for credit losses." If the model can accurately distinguish between a temporary financial dip and a structural decline in a borrower’s capacity to repay, the company could see a reduction in delinquency rates despite an expanded customer base.

Ultimately, the deployment of transformer-based underwriting is not just an incremental improvement to a checkout tool; it is a fundamental reconfiguration of Affirm’s risk-assessment DNA. As the fintech sector continues to converge with advanced AI research, Affirm is positioning itself as a data-first institution capable of reading the subtle signals of the modern consumer’s financial narrative, potentially setting a new standard for the BNPL industry at large. The coming year will be pivotal in determining whether this high-fidelity approach to credit can withstand the pressures of a volatile global economy.