Start with what the previous edition measured too narrowly, because it changes how the rest of this page should be read.

The first is the window. The previous edition said the RLTC margin improved 63 basis points over two years. On the two quarters chosen, that is true. On the next quarter's endpoints the same two years show −11.9 basis points. On a fiscal-year basis, +41.5. All three are arithmetically correct, which is the problem: “the margin improved 63bp over two years” reads as a property of the business when part of it is a property of the endpoints. FQ4'24 happened to print the highest quarterly margin until FQ3'26 matched it. Fiscal years provide the more stable basis — four quarters at each end, seasonality held constant, and the basis the company reports on. This edition therefore uses them as the primary comparison and shows all three windows.¹

On the fiscal-year basis, the central claim becomes more concentrated. Lower funding cost accounts for 39 of the 41.5 basis points of net improvement — a larger share than the 72% reported last time. The headline number became smaller, while the funding attribution became stronger.

The second correction is the growth attribution. The previous edition said the reacceleration came “mostly from acquisition.” That was overstated. On fiscal years, active-consumer growth went from +13.3% in FY2024 to +20.9% in FY2026 while transactions per consumer went from +26.6% to +20.0%, so the two legs are now roughly at parity. But the active-consumer leg outgrowing frequency is not the same thing as acquisition outgrowing frequency. Split that leg using Affirm's own first-time and repeat transaction counts, and the improvement appears much more on the retention/reactivation side than in first-time customer counts. Section four gives the full decomposition.

The third is the ticket extrapolation. The decline has been easing — −8.2%, −6.3%, −5.7% across the three fiscal years — and carrying a constant decline forward made the $100 billion arithmetic harder than the data warrants.

FY2026 closed at $50.2 billion of GMV, $4,261 million of revenue and $2,085 million of revenue less transaction costs, the last of which is 4.15% of GMV. Interest income is now roughly half of revenue, making lending economics central to how the business should be read.¹

Before turning to what needs watching, it is worth setting out what the year achieved. This is Affirm’s first fiscal year of positive GAAP operating income: $417 million against a loss of $87 million the year before, adjusted operating margin up from 24.1% to 29.0%, and an eleventh consecutive quarter of GMV growth above 30%.² Net income of $1.93 billion is a separate matter — most of it comes from a one-time release of the valuation allowance on deferred tax assets, an accounting event rather than cash the year earned, which is how the company describes it in its own letter.

One line worth watching sits further down the income statement. Gain on sale — the profit booked when Affirm sells loans rather than holding them — reached $597 million in FY2026: 14.0% of revenue and 28.6% of RLTC, up from 19.8% of RLTC two years earlier. On the fourth-quarter call the CFO explained that the line jumps in quarters carrying a non-consolidated securitisation — two in FY2026, a similar plan for FY2027, with no schedule announced in advance. Across the last eight quarters the line has swung between 21.5% and 34.1% of RLTC.^{1 3}

The accounting treatment is standard and fully disclosed. The modelling issue is timing. A line worth nearly three-tenths of RLTC that moves with deal timing can lift or depress any single quarter's margin. It matters for the same reason funding cost matters: both expose part of the reported unit economics to capital-markets pricing or transaction timing. The previous edition focused on the cost of funding; gain on sale is another part of the same picture, and it has become more material.

On a fiscal-year basis the RLTC margin went from 3.74% to 4.15% of GMV, an improvement of 41.5 basis points. It breaks down like this:

Contribution to the two-year change in RLTC as a share of GMV, fiscal-year basis, FY2024 to FY2026. The five parts sum to the net.¹

Cost lines improved 66 basis points over the two years: 39 from funding, 14.5 from credit provision, 6.8 from processing and servicing, 5.7 from the purchase commitment line. Take rate then gave back 24.5, leaving the 41.5 net. Funding is the largest single piece; the other 27 basis points came from operating lines Affirm can influence more directly. Funding is singled out here not because it is the only mover but because it is the input most directly priced by the market. It also reflects the company's credit performance, structure and execution, but the market still sets the price available at a given point in time.

Quarterly, the margin has oscillated between roughly 375 and 430 basis points since FY2025 began, with no clean trend, while funding cost has fallen steadily. That is precisely why the choice of endpoints mattered so much, and why this edition puts three windows on the page instead of one.

GMV is identically active consumers × transactions per consumer × ticket, and the identity is exact against Affirm's reported figures. Three fiscal years look like this:

The three legs of the GMV identity, fiscal-year basis. The last two columns are the absolute levels in that year.¹

Frequency has reached 7.00 transactions per active consumer a year, adding roughly 1.06 a year. That curve is not a pure measure of consumers using the original instalment product more often. The 10-K gives a narrower measure: Affirm Card was about 8% of total transactions in FY2024, 10% in FY2025 and 15% in FY2026, and the increase is accelerating.⁴ The company itself attributes part of the frequency gain to growth in Card active consumers, while noting that direct-to-consumer transactions earn lower merchant revenue. The 7.00 frequency measure therefore increasingly mixes repeat instalment use with higher-frequency Card behaviour. Those two activities have different unit economics, and public data does not separate them.

The active-consumer leg can be split one level further, and doing so changes the attribution. Affirm publishes two aggregate counts — first-time consumer transactions and repeat consumer transactions — so on the identity actives(t) = actives(t−1) + new customers − net lapse, the net additions from FY2022 to FY2026 decompose like this:

Net additions to active consumers. “New customers” is Affirm's disclosed first-time transaction count; “net lapse” is the residual implied by the identity.¹

Across five years the new-customer contribution has stayed between seven and eight million and has not accelerated with the platform. What changed is the other side: the implied lapse rate fell from 37.9% in FY2023 to 13.9% in FY2026. Roughly six-tenths of the improvement in net additions comes from there, not from more new customers.

Alongside flat new-customer counts, sales and marketing expense was $576.4 million in FY2024, $434.8 million in FY2025 (−25%) and $342.5 million in FY2026 (−21%), close to a 40% cut in two years.⁴ If acquisition spending falls that much while first-time transaction counts remain broadly stable, the pattern is more consistent with less reliance on paid consumer acquisition and more activity within the existing customer and merchant base. Public data cannot tell us how much of that shift was deliberate. The cut is not evenly spread either: in FQ4'26 alone, sales and marketing rose 16% year over year, which the company attributes to higher co-marketing expense and enterprise partner warrant expense — costs tied more closely to large merchant relationships than to paid consumer acquisition.

Put all of this against the constraint. Affirm indexes its entire medium-term framework to $100 billion of annual GMV and has never dated it. Carrying the ticket forward three years on the observed, easing decline gives about $234; at today's frequency the milestone then needs roughly 61 million active consumers, which is 86% of everyone Affirm says it has ever underwritten — 71 million — active in a single year.⁵ Management said at the Investor Forum that Affirm has transacted with more than half of those 71 million. Summing the workbook quarter by quarter, about 43.7 million people had transacted at least once by March 2026, so the statement holds and the true share is probably higher than “half” implies — but it is still a long way from 61 million. At ten transactions a year the milestone needs about 43 million actives; at twelve, about 36 million. Frequency has been adding roughly 1.1 a year, which puts ten inside the visible horizon.

The previous edition drew this constraint entirely on the consumer side, which was incomplete. Affirm is available at about 80 of the top 250 e-commerce sites and 10% of US e-commerce merchants, with merchant dollar-based net expansion averaging about 120% over three years and active merchants up more than 50% in the quarter.² Distribution is roughly a third penetrated at the top of the market, and the installed base can continue to expand volume without requiring a new merchant signing for every incremental dollar.

The other side of the same ledger deserves the same attention. Affirm disclosed in its outlook assumptions that a large enterprise merchant moved its pay-later volume to its own wallet, substantially completed during FQ1'26.² That raises a straightforward concentration question: how much of GMV sits in a small number of relationships? The 10-K answers part of it. The top five merchants and platform partners were 44% of GMV in FY2026, with Amazon alone at 22%, steady between 21% and 22% for three years. A second, independent exposure sits at origination: substantially all loans are originated by two partner banks, and the card is issued by two more.⁴

The wording needs narrowing here too, because stability alone does not establish weak network effects. Top-five concentration has in fact been drifting down — 47%, 47%, 44% — while Amazon has not moved at all. The same numbers are consistent with more than one explanation: network effects that remain strong at the head of the market, switching costs that make large integrations sticky, or some combination of the two. Public data cannot separate them.

The previous edition called underwriting the strongest pillar in the business, on the evidence that delinquency held a narrow band for four years while the book more than tripled. That judgement broadly stands, with two qualifications.

The first is that the band drifted. Thirty-day-plus delinquency excluding Peloton ran above the prior year in three of four quarters during FY2026 and closed the year at 2.5% against 2.3%; sequentially it improved, down 26 basis points from the third quarter. The allowance for credit losses was 5.9% of loans held for investment, up from 5.6% a year earlier and down from 6.0% in the third quarter.² The information runs both ways, and both directions belong on the page.

The second is how the line should be read. On the call the chief executive explained that the credit target is an input and the approval rate the output, described roughly 100 million credit decisions a quarter, and said you would see growth slow before you would see a credit disturbance.³ That is a useful description of how the system is managed, and it is also a modelling caution: when the portfolio-level loss rate is itself an input to the underwriting policy, stability at the portfolio level says a great deal about execution against policy. It tells us less, by itself, about the precision of the underlying model. Both matter, but they are different capabilities.

On the newest cohorts, Affirm has now published its own data, which is worth more than outside inference. Recent monthly instalment cohorts are tracking toward roughly 3.5% ultimate net charge-offs as a share of cohort GMV, which the company describes as in line with expectations and consistent with historic cohorts, and recent Pay in 4 vintages continue to track below 1% of GMV.² What matters now is where those cohort curves go over the next few quarters.

Taken together: the company held its loss rate inside a 2.3–2.8% band while growing quickly, let it drift up about 0.2 of a point in the most recent year, and its cohort-level ultimate loss expectations did not change. That still supports a relatively stable credit-performance reading. It can no longer be written as “delinquency didn't move.”

This is the part that has changed most since August, and it is where the analysis now turns.

Affirm's own outlook assumes rates rise. From the assumptions incorporated in the FY2027 outlook in the fourth-quarter supplement:

With that assumption in place the company still guides RLTC to about 4.16% of GMV, above the top of the 3.75–4% medium-term range it set for itself in May. That gives us a useful test for the coming year.

The external rate picture is broadly consistent with the assumption embedded in Affirm's outlook. As of the end of August 2026 the federal funds target range is 3.50–3.75%, held at the July meeting on a 9–3 vote in which all three dissenters wanted a quarter-point increase. The minutes record that “many participants assessed that policy tightening would likely be necessary if inflation did not decline.” Inflation has run above the 2% target for a fifth year, with supply shocks in energy cited directly; the June dot plot had nine of eighteen participants favouring at least one hike in 2026; and futures have moved from pricing cuts at the start of the year to pricing roughly 4% by year end.^{6 7}

So the easing that helped the last two years is largely behind the current numbers rather than sitting in front of them. Rates were cut into the current range and the Committee has since been on hold, with some risk now skewed toward a hike. The relevant question is no longer how much more benefit a future cutting cycle might provide; it is how much of the existing margin can hold if the rate backdrop becomes less favourable.

There is a second component to the tailwind, and much of the previous improvement has already happened. Affirm's funding cost is a benchmark rate plus a credit spread, and at the May Investor Forum the company disclosed that the spread on its static ABS deals had come in from roughly 300 basis points to roughly 100, a move management itself called a powerful tailwind.⁵ That 200 basis points of compression is already in the base. From roughly 100 basis points, the spread leg cannot repeat the same 200-basis-point improvement even if the benchmark cooperates.

That is what makes FY2027 useful. Affirm has supplied both a margin expectation and the rate assumption underneath it. If RLTC holds near 4.16% with benchmark rates rising and spreads no longer compressing at the same pace, that would provide stronger evidence that operating improvements can offset a less favourable funding backdrop. If the margin moves back toward 3.75–4%, funding conditions would still be a material part of the explanation.

Two things could also support the margin without requiring a corresponding improvement in the operating drivers, and both need to be separated when the year is scored. The first is the capital base already locked in: when the CFO discussed the FY2027 take rate he pointed at funding already brought on, which means some of today's cost advantage can carry forward even if new money becomes more expensive. The second is gain on sale, which at 28.6% of RLTC and swinging between 21.5% and 34.1% quarter to quarter can materially affect a full-year margin depending on deal timing.³

One more question deserves stating plainly, because the paragraphs above can otherwise make the funding exposure sound more dramatic than the balance sheet supports: if capital markets do tighten, how much buffer does Affirm have? Depending on the market for a price is one thing; having the supply cut off is another. As of 30 June 2026 the company held $2.6 billion in cash and available-for-sale securities, $5.3 billion of undrawn secured funding capacity and a fully undrawn $675 million revolver — roughly $8.6 billion of available liquidity against average funding debt of $7.5 billion for the year — and the first maturity wall on its secured debt does not arrive until FY2028 and is not large. Separately, the company reported funding capacity of $30 billion at the end of the quarter, which on its own estimate supports more than $70 billion of annual GMV against a FY2027 guide of $64 billion.^{2 4} The one caveat is that the $5.3 billion is capacity against pledgeable loan assets rather than unconditional cash, so advance rates falling in a credit downturn would make the usable figure smaller than the stated one.

One result from the scorecard deserves separating out, because it says something about Affirm rather than about the forecast.

The forecast being scored was not built on private access or a proprietary view. It was built on Affirm's own published methodology — the June 2023 Financial Model Information Session, where the company set out how to model each revenue line against its correct denominator — applied to Affirm's own published historical financials. Every input was public.⁸

That model reproduced the company's forward growth rate to within a tenth of a percentage point and its FY2027 margin guide to within about 1.6 basis points. The more useful point is how much of the forward framework can be reconstructed from the information Affirm has already published.

Guidance often contains assumptions that are difficult to inspect from outside. Here, much of the framework can be rebuilt using the company's own stated method. That is a positive statement about disclosure quality. It also makes the forward numbers unusually checkable: a reader can reproduce a large part of the logic rather than treating the guidance as a black box.

This is the last regular edition, so what follows is written to be scored without us. Two reviews are planned: one around February 2027 after the second-quarter print, and one after FY2027 closes.

As of 8 September 2026, nothing has been checked. Publishing it now rather than later is the point.