The Buy Now, Pay Later (BNPL) sector has long been characterized by a tension between growth and risk management. As Affirm scales its operations, the company has deployed a new transformer-based underwriting model across its United States checkout infrastructure. This technological pivot represents a significant departure from traditional credit scoring methods, moving away from static snapshots of financial health toward a dynamic, time-sensitive analysis of consumer behavior. By leveraging transformer architecture—the same underlying technology that powers advanced generative artificial intelligence—Affirm aims to decode the sequence and cadence of financial events to better predict creditworthiness.
The Technological Shift in Underwriting
For decades, the financial services industry has relied heavily on the FICO score and similar metrics as the primary barometer for credit risk. These scores condense a consumer’s complex financial history into a single, three-digit number. While this provides lenders with a high-level overview, it often lacks the granular context required to assess individuals whose financial situations are evolving.
Affirm’s new model addresses this "compression problem" by analyzing the trajectory of a consumer’s credit history. According to company leadership, the model is specifically designed to interpret the timing and sequence of credit events. For instance, two consumers might have identical credit scores, but one may be demonstrating a pattern of financial recovery, while the other shows signs of increasing instability. Conventional models often treat these two individuals as identical risks. Affirm’s transformer-based approach identifies the nuances in these trajectories, allowing the company to approve applicants who might have been rejected under the constraints of legacy systems.
Impact on Transaction Volume and Risk
The initial deployment of this technology has yielded measurable improvements in Affirm’s operational efficiency. The company reported that the new model facilitated a 3.4% increase in completed purchases compared to a control group operating under previous underwriting standards, all while maintaining a comparable risk profile.
This improvement is critical for Affirm, as the company has evolved from a niche point-of-sale financing firm into a comprehensive financial ecosystem. Its product suite now includes the Affirm Card, a dedicated consumer app, a marketplace, and the Affirm Money Account, a high-yield savings product. This diversification has led to a surge in user engagement. In fiscal year 2026, Affirm reported 27.8 million active consumers—a 21% year-over-year increase—with the number of transactions per active user rising from 5.8 to 7.0. As the frequency of smaller, everyday purchases increases, the accuracy of the underwriting engine becomes the primary driver of profitability and risk containment.
Chronology of Affirm’s Strategic Evolution
Affirm’s path toward AI-driven lending has been marked by a consistent focus on data integration and platform expansion:
- 2012: Affirm is founded with a focus on transparent, installment-based point-of-sale lending, aiming to disrupt traditional credit card models.
- 2020-2021: The company expands its product footprint, launching the Affirm app and beginning the integration of direct-to-consumer shopping features.
- 2022: Affirm deepens its relationship with retail partners, moving beyond luxury goods into everyday essential categories.
- 2024: The company begins testing transformer-based machine learning models to analyze non-traditional data points, moving away from reliance on static credit bureau data.
- 2025 (Q4): Affirm officially deploys the transformer-based underwriting model across its entire U.S. checkout, marking a new era in their proprietary risk assessment capabilities.
- 2026 (Fiscal Year): Affirm achieves significant scale, reaching 27.8 million active consumers and demonstrating the efficacy of its high-fidelity underwriting models in managing larger transaction volumes.
Insights from the C-Suite
The shift toward transformer architecture was underscored in recent public commentary by Affirm President Libor Michalek. During an industry podcast, Michalek noted that traditional models suffered from a lack of "fidelity" when mapping the time dimension of credit events.
"The time dimension of multiple purchases, multiple credit events in a customer’s life wasn’t being represented with particularly high fidelity," Michalek stated. By "fidelity," Affirm implies that previous machine-learning iterations were effectively blurring the signals that indicate a consumer’s intent and ability to repay. By upgrading to transformer-based models, which excel at identifying long-range dependencies in data sequences, the company believes it has gained a superior vantage point into consumer financial health.
Fact-Based Analysis of Market Implications
The transition to advanced transformer models has broader implications for the fintech industry. By successfully increasing approval rates without increasing the default rate, Affirm is signaling that traditional credit bureaus may be providing insufficient data for the modern economy.
There are three primary implications for the sector:
- Increased Competition: As Affirm improves its ability to identify "hidden" prime borrowers among the subprime or thin-file populations, other fintech players will likely accelerate their own investments in proprietary AI models. The competitive moat for BNPL firms is no longer just merchant partnerships; it is the quality of their data science.
- Regulatory Scrutiny: As algorithms become more complex, regulatory bodies such as the Consumer Financial Protection Bureau (CFPB) are likely to increase their oversight of "black box" underwriting. Affirm will need to ensure that its transformer-based decisions remain explainable and compliant with fair lending laws, which prohibit discrimination based on protected characteristics.
- Customer Lifetime Value: By identifying consumers who are on a positive financial trajectory, Affirm can cultivate long-term loyalty. If the company can capture a user during a period of financial growth, it increases the likelihood that the consumer will utilize the broader suite of Affirm products, including savings accounts and debit-style products, effectively deepening the relationship and increasing customer lifetime value.
Challenges and Future Outlook
Despite the success of the initial rollout, the road ahead is not without obstacles. Transformer models require vast amounts of compute power and high-quality data to function optimally. As Affirm expands, maintaining the model’s accuracy in the face of shifting macroeconomic conditions—such as inflation, changing interest rates, and employment fluctuations—will be paramount.
Furthermore, the integration of multiple product lines presents a unique data challenge. Affirm must harmonize the data collected from its savings products with the data from its checkout lending products to ensure a holistic view of the consumer. This requires robust data architecture and continuous model retraining to prevent "concept drift," where the model’s predictive power wanes as market realities change.
The company’s focus on "inward" growth suggests a strategic maturation. Rather than chasing raw customer acquisition at any cost, Affirm is prioritizing the optimization of its existing base. By leveraging advanced AI to better understand the behavior of its 27.8 million users, the company is positioning itself to be more than just a payment option at checkout; it is aiming to be a central pillar of the consumer’s financial management strategy.
In summary, Affirm’s deployment of transformer-based underwriting is a calculated technological bet that the future of credit lies in the analysis of behavior over time. As the firm continues to scale, its ability to translate this complex data into consistent, safe, and efficient lending decisions will likely be the definitive factor in its long-term financial performance. The company’s ability to outperform its previous machine-learning models while expanding the pool of approved borrowers demonstrates that even incremental gains in model fidelity can lead to substantial shifts in growth and profitability. As the BNPL market matures, Affirm’s data-first approach serves as a blueprint for how financial institutions can bridge the gap between traditional credit metrics and the fast-paced, high-frequency nature of modern digital commerce.
