
Shares of Affirm Holdings Inc. (AFRM) gained about 0.42% in the overnight session late Sunday, shortly after the company launched a transformer-based machine learning model for real-time underwriting.
Affirm CEO Max Levchin said in an interview with CNBC that while human underwriters “could never do what we're doing with transformers,” they would not replace them completely.
Affirm’s transformer-based underwriting model, which went live in the U.S. last week, is built on 14 years of transaction-level data and underwriting experience. The model analyzes the sequence and timing of events across a consumer’s credit history, helping identify patterns that traditional models may miss.
The system is designed to initially approve additional eligible applicants, including consumers with limited credit histories or no FICO scores. Its own research shows that compared with a control group, it generated 3.4% more completed purchases, while the additional loans performed better than a comparable expansion under Affirm’s previous models.
Affirm said the model maintains explainability and real-time speed while enabling more responsible credit approvals.
Levchin said in the interview with CNBC last week that the company’s new transformer-based underwriting model is designed to augment, rather than replace, human judgment.
He said that the model could parse buckets of information with different numbers of dimensions, and that ultimately “a human decides what's worth looking at, what's not worth looking at.”
The CEO said traditional machine learning essentially builds a decision tree by identifying specific factors, automating how they are evaluated, and repeatedly sorting borrowers into categories based on perceived risk until the final outcome is a simple yes-or-no decision.
The model can also analyze credit information at a level traditional underwriting systems cannot, including details buried in credit reports that may provide important context about a borrower’s circumstances. Levchin said the model can combine simpler decision-tree models with “extremely multidimensional models” to uncover deeper signals in a consumer’s credit history.
He gave the example of a borrower explaining a late payment in a letter, saying, “The reason I was late is [that] because I changed my address and the bill didn't reach me in time.” A traditional underwriting model would likely discard such information, Levchin said, while the transformer-based system can incorporate it and determine whether it changes the borrower’s situation.
In a post on X, Levchin also clarified that the system would not pose any threat to humans, as concerns over AI use have been growing in the market.
“Also, for the avoidance of doubt, this particular foundational model has no means of “breaking out” or harming its fellow humans or agents. It’s got one purpose, and one purpose only: to help you, the borrower, make the smartest financial decision every. single. Time,” he said.
AFRM stock is up more than 60% in the last six months. The stock gained about 1.32% after its AI underwriting model announcement.
On Stocktwits, retail sentiment around AFRM stock was ‘bearish’ at the time of writing.
One user said last week, “$AFRM is going to 65, trust me , i know what i m saying.” The company's shares closed at $71.51 on Friday.
However, another bullish user said, “$AFRM potential. potential potential potential. Candidates for next index rebalancing 100%.”
AFRM stock is down nearly 21% in the last year.
For updates and corrections, email newsroom[at]stocktwits[dot]com.<
Stay updated with all the latest Business NewsShare Market NewsIPOsGold PriceDA Hike8th Pay CommissionAsianet News Official App