Affirm Boosts Approval Rates with New Transformer-Based AI Underwriting Model

Affirm has launched a transformer-based underwriting model in the U.S. to analyze the sequence and timing of consumer credit events in real-time. This shift from traditional machine learning to deep learning architectures allows the lender to approve more customers, particularly those with thin credit files, while maintaining strict risk controls at the point of sale.

What was announced

Affirm has introduced a new transformer-based model to its underwriting engine, marking a significant technical shift in how the company evaluates creditworthiness at the point of sale. Unlike previous iterations that relied on static summaries of credit data, this architecture is designed to interpret the specific order and timing of events within a consumer’s credit history. The model is currently live for consumers using Affirm at checkout across the United States.

In its initial deployment phase, the model demonstrated a measurable impact on conversion and risk management. Affirm reported a 3.4% increase in completed purchases compared to a control group using its previous machine learning models. Crucially, the system was able to approve applications from individuals who would have otherwise been declined, including those with limited credit histories or no FICO scores. Despite the broader approval criteria, these additional loans performed better than comparable expansions under the company’s older underwriting frameworks.

The technical implementation required the development of a proprietary algorithm to solve the dual challenges of speed and explainability. Because Affirm provides real-time decisions at checkout, the transformer model must process complex sequential data instantly. Furthermore, the company has integrated validation and monitoring tools to ensure the model provides the same level of transparency and "explainability" as traditional machine learning, which is a critical requirement for regulatory compliance in consumer lending. The system continues to utilize credit-bureau measures such as balances, utilization, and payment history, but identifies patterns within them without needing pre-designed measures for each pattern.

"Underwriting is the heart of what we do. The goal isn’t to approve every transaction, it’s to make the right decision for each one. We don’t benefit from extending credit that can’t be repaid, which means saying yes to more people only works when we get even better at saying no."

Libor Michalek, Affirm President.

The companies involved

Affirm, listed on the NASDAQ under the ticker AFRM, has spent 14 years positioning itself as a transparent alternative to traditional credit cards. The company’s core value proposition revolves around its "no late or hidden fees" policy and its practice of underwriting every individual purchase in real-time. This approach differs from the revolving credit lines offered by banks, which typically assess a borrower’s creditworthiness once and then allow them to spend up to a limit. By making a unique decision for every transaction based on what a user can responsibly repay at that specific moment, Affirm has carved out a significant share of the U.S. e-commerce market.

The San Francisco-based fintech has become a dominant force in the Buy Now, Pay Later (BNPL) sector, leveraging in-house machine learning models to manage risk. The company’s leadership, including President Libor Michalek, has consistently prioritized the refinement of these proprietary models as their primary competitive advantage. Affirm remains one of the most scrutinized and influential players in the intersection of artificial intelligence and consumer finance, with its technology integrated into thousands of merchant checkouts to facilitate point-of-sale financing.

What FF News has reported before

FF News has closely followed Affirm’s integration into various sectors of the global economy. In September 2026, we reported that Flex Dental Solutions Launches BNPL and Surcharging to Boost Practice Revenue, highlighting the expansion of Affirm’s services into healthcare. This followed an August report on how Airwallex Integrates Affirm to Enable Buy Now Pay Later for Global Merchants Selling in the US, a move that broadened the lender's reach to international sellers.

The broader market context has also been a focus of our coverage. We recently analyzed data showing that SMEs Embrace Flexible Payments to Compete with Retail Giants as BNPL Demand Surges. However, the rise of these services is not without friction; FF News also covered a Chargebacks911 report titled BNPL Fraud Risk: 40% of Merchants Warn of Rising Dispute Complexity in New Chargebacks911 Report, which underscored the operational challenges facing merchants in the BNPL space.

What this means

The adoption of transformer models—the same architecture behind modern generative AI—represents a major evolution in credit risk assessment. By moving away from static credit bureau snapshots and toward sequential data analysis, Affirm is challenging the traditional dominance of the FICO score. This move puts significant pressure on legacy lenders who rely on older scoring models that often fail to capture the nuances of "thin-file" borrowers. The industry is now entering an era where the ability to find "signal" in existing data is more valuable than simply acquiring more data. However, as these models become more complex, the sector faces an ongoing challenge: maintaining the "explainability" required by regulators while pushing the boundaries of predictive accuracy.