Affirm Holdings (AFRM) climbed 5.5% in premarket trading Thursday after the buy now, pay later company unveiled a new transformer-based artificial intelligence model designed to deliver faster and more accurate loan decisions at checkout. Shares had been trading around $71.58 heading into the announcement, well below the average analyst price target of $98.78.

Buy now, pay later services allow consumers to split purchases into installment loans, with approval decisions issued in real time as transactions are completed online. Because underwriting happens while the shopper is still mid-purchase, changes in decisioning show up directly in checkout outcomes such as completed purchases.

AI Model Now Live at Checkout

The new model is already live in the United States and draws on 14 years of Affirm's own transaction-level underwriting data. It applies machine learning to analyze the order and timing of events in a consumer's credit history, which the company describes as a step up from its previous approach.

Transformer architectures process sequential data, making them well suited to reading ordered event histories; the same class of model underpins modern large language models.

One notable aspect of the system is how it handles thin-file borrowers. Affirm said the AI approved additional applicants that its older system would have turned down, including consumers with limited credit histories and no FICO scores. That reach beyond conventional scoring is where reliance on Affirm's own first-party transaction data, rather than external bureau files alone, becomes a practical advantage.

Stronger Results With Lower Risk

In its initial rollout, the model produced 3.4% more completed purchases compared with a control group. Affirm also said the additional loans performed better than a comparable expansion run under its previous machine learning models, suggesting AI is selecting better risks rather than simply approving more borrowers. For a lender, that distinction matters: approval growth only adds value if the incremental loans are repaid.

Libor Michalek, Affirm's President, said the transformer architecture allows the company to extract new information from data it already holds. “Seeing a credit history more clearly means we can responsibly say yes to more people,” he said.

The model identifies patterns within and across credit accounts and tracks how those patterns shift over time. Affirm built a proprietary algorithm to keep the model explainable while remaining fast enough to run in real time at checkout — a practical constraint for a lender issuing automated credit decisions in seconds.

The company stressed that the goal is not to approve every application, noting that extending credit to consumers who cannot repay does not benefit the business.

Insider Selling Draws Attention

Chief Accounting Officer Siphelele Jiyane sold 26,980 shares on September 14 at an average price of $72.19, generating roughly $1.95 million. The transaction reduced his ownership by 12.51%, leaving him with 188,711 shares. Jiyane had also sold 25,000 shares on September 4 at $72.41. The back-to-back sales are likely to attract scrutiny, even though insider selling does not always signal a negative outlook.

Earnings and Analyst Coverage

Affirm reported EPS of $4.62 in its most recent quarter, well above the $0.35 consensus estimate. Revenue came in at $1.07 billion, up 33.1% year over year, though slightly below the $1.11 billion estimate.

Analyst reaction has been broadly positive. Piper Sandler raised its price target from $103 to $115 while maintaining an overweight rating. BMO Capital Markets and Citigroup both reiterated outperform ratings. Of the 32 analysts covering the stock, 24 rate it a Buy and 8 a Hold, producing a “Moderate Buy” consensus with an average price target of $98.78 — well above the recent price of $71.58.

Affirm's 50-day moving average stands at $75.43, while the 200-day moving average sits at $66.39, placing the current share price between the two. As the model processes more checkout volume, its early benchmarks — the 3.4% lift in completed purchases and the stronger performance of incremental loans — are the figures to compare against future reported results.