Affirm Holdings fiscal Q1 2027 earnings call, November 5, 2026

A transformer model that finds better risks without more losses

The most substantive new disclosure on the call was technical. Founder and CEO Max Levchin described a transformer-based underwriting architecture that learns repayment patterns and generates candidate features without human involvement, then trains on them in a single pipeline. Historically, Affirm's data scientists hypothesized patterns, hand-coded them into features, and screened them for any correlation with prohibited bases such as race or location. Levchin said the new approach "picks up information value that may be just too subtle for a human to detect," and does so much faster.

The quantified claim is the one investors should focus on. In Affirm's standard head-to-head test of a new candidate against the incumbent model, the transformer approach separated good risk from bad "twice as well" as the next candidate in a model family the company has trained for nearly 15 years. Levchin was careful about what that means in practice. It does not mean twice as many approvals. It means roughly 2 to 3 points more approvals, made "essentially with no incremental loss whatsoever." He framed it as a second growth lever alongside new products: finding better risks without taking on more delinquencies. He added that more models in the family are in the pipeline. The claim comes from management and is not independently verifiable, though the company has published a roughly 2,000-word technical blog post that Levchin said is "pretty disclosive" apart from trade secrets.

Provisions rose on deliberate experiments, and the market noticed

The stock was under pressure after the print, and Levchin's opening quip about "a blue tsunami on Tuesday and a red wedding today" acknowledged as much. Several analysts pointed to higher provision expense as a share of GMV and a modest rise in delinquencies. CFO Rob O'Hare gave a specific, self-inflicted explanation. Affirm ran experiments in approvals and originations, localized to September, and chose to keep those loans on its balance sheet. Because they are experimental, they were reserved conservatively, which pushed up allowance and provision expense in the quarter.

O'Hare argued the trade was worth it, pointing to a revenue less transaction cost take rate of 4.3% in the quarter and saying the experiments were accretive to the network, particularly for new user acquisition. Levchin dismissed the year-over-year delinquency change as seasonal noise. Both emphasized the short weighted average loan term. "We get to deal with whatever overexposure we might take on today in less than 90 days," Levchin said. President Michael Linford added that an asset-backed securities deal priced on the day of the call was more than 4x subscribed despite what he called one of the worst equity capital markets days since COVID. He cited this as evidence that sophisticated credit buyers trust Affirm's loss control.

The honest read is that management is deliberately spending credit dollars to buy growth and data, and that the evidence for the payoff rests on the take rate and on investor demand for the paper. The market's discomfort with that choice is understandable, even if the short duration limits the downside.

Growth: a 30% floor and a high-30s start to Q2

Fiscal 2027 guidance now carries a GMV growth floor that "starts with a 3," in O'Hare's words, meaning at least 30%. Q1 GMV growth was 34%. Management said growth in September and early Q2 ran "much north" of Q1 and that the first month of the second quarter is growing at a high-30s rate. Asked whether a floor of 30% implies deceleration, O'Hare said there was nothing specific to point to and that the floor is a responsible setting, not a range with a ceiling. Levchin argued Affirm is under 4% of e-commerce and live in just 4 U.S. markets, so growth is "ours to screw up."

GAAP margin guide: warrant expense is the gap

Analysts noted that the Q2 GAAP operating margin guide came in below consensus. O'Hare attributed most of the gap to warrant expense tied to a large partner program in which the partner earns warrants for new-to-program users. He said the Street is probably under-modeling it. New users are growing quickly through that program, partly because of international expansion, including a toehold in the U.K. Management sees this as good news for growth and a drag on GAAP margin, and it expects holiday seasonality plus the higher new-user mix typical of a new market to shape the expense.

Funding costs: sequential uptick, fixed-rate cushion

Funding costs rose 27 basis points sequentially, which Autonomous Research noted is already well above the old rule of thumb of about 40 basis points per 100 basis points of benchmark rate moves. O'Hare said not to read the increase as a pure pass-through of rising rates. Funding costs as a percentage of loans are still down year over year at about 6%, more than 70% of on-balance-sheet funding is fixed rate, and Affirm has recently issued 3-year and 5-year deals. He said this visibility is reflected in guidance.

Direct-to-consumer mix dilutes merchant revenue but not economics

Direct-to-consumer and wallet activity now drives about half of volume, according to Goldman Sachs. Jefferies asked about lower merchant revenue relative to GMV. O'Hare confirmed that card and virtual card transactions earn interchange, which is generally lower than the merchant discount rate on integrated merchants, so network revenue is under some pressure. The offset is mix: Affirm Card has historically run above 80% interest-bearing loans, so total compensation per dollar is higher.

Levchin described the card as a "trailblazing device" to pull merchants into direct integrations, after which 0% and other subsidized programs appear on the card as well. About 30% of card transactions are in-store, plan creation before the swipe remains the dominant behavior, and a smaller share of users convert purchases into plans after the transaction. He said card-specific financing programs and card-only offers will launch "in days, if not weeks," with a next-generation card in development at the zt1 team.

zt1 and brand-sponsored promotions

Levchin gave a rare look at how the zt1 incubation team operates. Projects must show a risk-adjusted path to $1 billion of revenue within 5 years of launch, or move an existing $1 billion business by 15% to 20%. The one example he detailed was brand-sponsored promotions. General merchants typically mark up branded goods by only 10% to 15%, so they cannot fund a 0% offer. The manufacturer, however, retains margin and marketing budget. Affirm is building software that lets brands fund 0% financing on specific SKUs at full retail price, replacing the spreadsheet-and-phone-call process of traditional vendor-funded promotions. Levchin said about 1 million SKUs in the Affirm app are already brand-funded and that he hopes to reach "hundreds of millions" over time, while asking investors to "discount heavily" his enthusiasm.

Agentic commerce: favorable positioning, but early

Levchin argued that agents will optimize for the most rational pricing per consumer, not simply 0%, since many lower-income users prioritize a monthly payment under $100 over a headline APR. He called Affirm "the only game in town" that prices each transaction individually in real time. He told Mizuho that merchants should subsidize more Affirm 0% offers to win agentic conversion, and that Affirm is building programmatic APIs so that merchants' agents can deploy promotional dollars to buying agents. He expects 0% volume to rise. He also tempered expectations, noting how few agentic rides Uber has booked, and cautioned against building "a spreadsheet that shows going to the moon."

On risk, Levchin was blunt that early agentic commerce will be "a little bit messy," with hallucinated purchases, returns and agentic fraud all likely. He framed that as an opportunity, since Affirm's edge is in data and ML-driven risk management, and he said attackers and defenders are both gaining AI capability but he believes defenders are getting access to better models faster. He acknowledged a current "shooting war" of daily vulnerability disclosures.

Services and large tickets; Edge remains distant

The services vertical, which includes Intuit and ServiceTitan partnerships, is the fastest-growing segment Affirm breaks out. Levchin said it opens home improvement and offline services, but each new vertical takes several quarters to reach product-market fit. On ticket size, he said the largest transaction Affirm wrote was once $7,000 and that the company is now well beyond that, with a cautious expansion of risk appetite for high-credit-quality borrowers. He could not say whether merchant discount rates would rise or fall. Merchants selling a $20,000 to $29,000 HVAC system will pay up for conversion, but low-risk buyers need less margin of safety. O'Hare noted that longer durations lift total revenue per loan, offset by more funding cost.

Affirm Edge, the debit-issuer distribution product, disappointed anyone looking for specifics. Pilots are slated for the first half of calendar 2027, but Linford said there was "nothing to announce" and that "patience is the code word," given how slowly bank and platform partners move.

Operating savings

Levchin flagged ACH-based down payments as one example of continuing cost reduction. Moving off debit networks saves interchange but introduces non-real-time settlement risk, which Affirm believes it can cover with a specialty underwriting model. He said the savings go toward better consumer deals and lower operating expenses, but gave no quantification.