Fair Isaac

Fair Isaac makes the FICO score, the three-digit number between 300 and 850 that American lenders use to judge whether you will pay them back. A ten-year-old could follow the idea. Three credit bureaus, Equifax, Experian and TransUnion, keep a file on nearly every adult: what you owe, whom you owe it to, whether you paid on time. FICO wrote the recipe that turns a file into a number. It owns the recipe and none of the ingredients. Each time a lender asks for a score, a small fee flows back to FICO.

That fee was tiny for a very long time, and the cost of running the recipe once more is close to nothing. So the Scores business earns a margin of about 91%. Scores brought in $1.65bn of the $2.53bn of revenue in the fiscal year to 30 September 2026, about 65%, and on the last segment figures we have, in the June quarter, it made nearly 90% of segment profit. Within Scores, the money that matters is a single line: mortgage originations. That one line was about 42% of company revenue in the June quarter and about 37% in September.

The rest of the company is Software: decision and fraud tools for banks, sold under the FICO Platform name. It had revenue of $883m last year at a margin in the mid-20s. It is a respectable business that nobody bought the stock for, and it now matters more than it ever has.

The whole enterprise needs very little capital: $38m of capital spending last year, 1.5% of revenue. It has turned that fact into a remarkable financial structure. It has spent roughly $9.5bn on its own shares since 2012, has cut the share count by about two-fifths, and today has $5.26bn of debt and negative book equity of $3.87bn.

Add one more cell in your head. Over five years, FICO has returned +62% against +80% for the index. A shareholder who bought in late 2021 has spent five years, through the greatest price increase in the company's history, to finish behind a fund that charges almost nothing. At its record close of $2,382.40 on 26 November 2024 the stock was worth about 66 times what it fetched when Will Lansing became CEO in January 2012. It is now worth about 19 times. That is still a wonderful result, and it tells you how far the price had run ahead of everything else.

The neatest summary of this history is that the franchise began with a letter from Freddie Mac and was opened up by posts from Freddie Mac's conservator, and that FICO's consent was required for neither. Everything impressive that Lansing did sits between those two events. He saw that a product worth a great deal to lenders was being sold for almost nothing, and he had the patience to rewrite the contracts. He also never built a defence against the one party that could change the rules, and it is hard to say what that defence would have been. The company's own 10-K has long named the risk: if other credit score models are approved, it could have a material adverse effect on revenue and on the stock. Few people read that sentence until 2025.

We can explain how FICO makes money in a paragraph, which is the first test. The second test, whether we can say what it will look like in ten years, is where the trouble starts. Until 2025 this was the sort of business you could leave alone for a decade. Today the forecast depends on what a political appointee does next, on how three companies that are both customers and competitors choose to price, and on private contracts between FICO and individual lenders whose terms are not public. Those are hard things to know from the outside.

The ladder is not quite what the headlines say. Politicians quote a rise from 60 cents to $10, and it is a real rise, but 60 cents was a royalty paid by the bureaus and $10 is a wholesale price into resellers. FICO's defence is that $10 equals what the bureaus used to charge on average, since they marked up its product by about 100%, and that its own cost to a lender did not change. The revenue line settles the argument. In the June quarter, mortgage origination revenue rose 97% while volumes rose in the low single digits. FICO's take per score about doubled, and the mark-up moved from the bureaus to FICO. Both statements can be true: no increase to the lender, and twice the money to FICO.

You can see who paid in the bureaus' filings. TransUnion's US margin fell 230 basis points in its third quarter, which it put down mainly to FICO's royalties growing faster than revenue. Equifax's US mortgage revenue rose 15%, but 1% without the FICO pass-through. So the bureaus have been carrying a product that shrinks their margin, sold by a supplier trying to go around them. Anyone looking for the reason they are prepared to give VantageScore away should start there.

Is it expensive? It depends on what you hold it against. FICO says its contribution to a tri-merge bundle is at most $30, on a mortgage that costs thousands to close. In 2024 Lansing argued that royalties were about two-tenths of one per cent or less of closing costs. As economics, that is a sound argument. It stopped working when a rival arrived at one-tenth of the price, and when Senator Hawley and Mr Pulte decided that the size of the fee was less interesting than the size of the increase.

Taken alone, these are fine results. Revenue came in slightly above the roughly $652m that the full-year guide implied, and non-GAAP EPS of $42.99 beat the $42.43 guide. Software revenue grew 19%, after 2% in June. Free cash flow passed $1bn for the first time. Debt fell by about $320m in the quarter, and there were no cash buybacks at all: the $3,046m spent in the year had already gone by June. After the borrowing of the first nine months, that is the first sign of sobriety.

Then comes the sentence. Under "Outlook", FICO writes that it is "transitioning to bespoke mortgage originations pricing, tailored to the individual needs of our customers", that it is "in the middle of this pricing transition", and that "we are not providing fiscal year 2027 guidance at this time." A company that gave a guide for 2026 and raised it in July now declines to say what next year looks like. There are honest reasons for that. Deals are unfinished and the mortgage market is weak. But think about what the words mean. For eight years FICO told the market what a score would cost, and the market wrote it down. The word "bespoke" means that there is no longer one price. The toll has become a negotiation, conducted lender by lender, and the gatekeeper does not yet know what the gate will collect. In thin trading in the first fifteen minutes after the release, the stock went from about $740 to about $630, some 10% below the close, and after the call it recovered to about $658, some 6% below. Treat those as first reactions only. At $658 the price would still be about 16% above our central value, and well above the level at which we would become interested.

The settlement is the night's good news. The release discloses a $65m accrual for "a proposed settlement of the antitrust class action lawsuits related to our Scores business". This is the case filed in 2020, consolidated before Judge Chang in Chicago, in which the claim under section 2 of the Sherman Act survived dismissal, and in which damages can be trebled. Nobody outside could put a number on that risk. If the court approves this, a suit that hung over the stock for six years will have cost about $3 a share, about three weeks of free cash flow. We would like to see the terms in the 10-K, especially any change to the contract clauses the plaintiffs attacked. It does not touch the Florida Attorney General's demand or the Senate inquiry, and it says nothing about what the Preferred Program may invite in future. Still, it removes a large unknown at a low price.

Three smaller things deserve a harder look.

First, volume. The release says B2B growth of 39% came from price, "partially offset by a decline in mortgage originations volume". In the June quarter volume was up. How much of the fall was the market, and how much was lenders dropping FICO, is the most important question in this report, and the CFO answered it on the call: volume fell by a high single digit on a year earlier, which he put down to high interest rates, and no meaningful volume went to competitors. With mortgage revenue up 72%, that means the price per score rose by about 85%, on our arithmetic. We accept this for the quarter, while noting that it is the seller's account. Scores revenue still fell about 10% from the June quarter, against about 4% a year earlier. One grid only began on 1 October, so these numbers do not yet show its effect. The first quarter that can is the one now under way.

Second, the cost of the debt. Net other expense of $29m is flattered by a $48m gain on an equity stake. Strip that out and it ran at about $77m a quarter, or about $310m a year, against $122m in all of fiscal 2025. Interest is taking about a fifth of operating profit before the one-offs.

Third, the one-offs. Non-GAAP net income of $239m excludes the settlement, $27m of restructuring and $44m of stock pay, and removes the $48m gain. Taking out the gain is honest. Taking out stock pay, as ever, is not something an owner should do. Selling, general and administrative costs rose from $126m to $225m; even if the whole settlement sits in that line, as we assume, they rose 28% before a 15% cut in positions takes effect. The CFO said that much of that saving will be reinvested, and gave no figure.

What the call added. Analysts pressed on everything the release left out, and they got a few facts and several refusals. Asked whether the average price of a FICO score will go up, down or sideways, Lansing said: "We're not going to answer that." He would not give the uptake of the Preferred Program, the other programmes on offer or the share of volume under each. The CFO gave no range for EPS. A guide will come once more of the pricing is settled, with no date. Rocket, UWM and Mr Pulte were not named at all. Lansing repeated that the gaming between the two scores is real, argued against a single grid, and said he has seen no share lost to Vantage in other lending. We would not read too much into one refusal, because these are live negotiations. But a company that had lost 62% of its value, and expected a higher average price, would have good reason to say so. That is why our central case still has mortgage revenue falling about 3% a year.

In May 2010 Warren Buffett told the Financial Crisis Inquiry Commission that the single most important decision in evaluating a business is pricing power. Then he gave the test, talking about Moody's, another franchise wired into financial regulation. If you can raise prices without losing business to a competitor, you have a very good business. And "if you have to have a prayer session before raising the price" by a fraction of a cent, you have a terrible one.

For eight years FICO was the finest illustration of that sentence in American business. Ask what happened in October. The price decision for 2027 needed a named programme, a bundle with a second product, a commitment asked of the customer and a statement from the CEO about homeownership. That is a prayer session. It is also, at a segment margin of 91%, a very long way from a terrible business. The moat must be taken apart, because its pieces are moving in different directions.

What does each party really depend on? Lenders depend on whichever score the GSE grid prices, and with one grid the better number wins. Vantage scores tend to run higher: about nine points on average on Fannie loans in one study, with many buckets at 15 to 20. So a lender who pulls both can offer some borrowers a cheaper loan. Rocket says it tested 1.4m reports and that those who benefited saved $1,600 on average. Investors in agency mortgage bonds hold a government-backed guarantee, and they care about prepayments far more than default. A Santander note in September expected little effect on valuations, which undercuts FICO's claim that Vantage loans will trade worse. The dependence is real among those who hold credit risk: mortgage insurers, who must hold more capital on Vantage-only loans, and credit-risk investors. And there is a warning from Fannie Mae itself, whose latest 10-Q says the move could change the credit quality of what it buys and lower its guaranty fees.

Our verdict is that a government-conferred toll has become a commercial standard that must defend itself. That is still a business, and possibly a very good one. But it is no longer the business that traded at 40 times forward earnings, and a moat that a regulator can drain by post should be valued differently from one that he cannot reach. We score it 6 out of 10, down from what would have been 9 two years ago.

The fall on 26 October is the one to think about. FICO announced that it would hold the price flat at $10 for lenders who commit to it, in a bundle with its newer score, 10T. It reported no bad number. The stock fell 13%, because the market had capitalised a rise every year for ever. Mizuho had modelled only about 5% for 2027, after a year in which FICO's take doubled. So a modest assumption went, and with it the multiple.

The Preferred Program, as far as anyone can tell. $10 wholesale per score, flat on the year, sold through participating resellers, with Classic and 10T together. Reports from Reuters and HousingWire say lenders commit to pull FICO on 100% of files. That condition is not in the release. It is reported to be non-exclusive, so a lender may add Vantage but not substitute it. It needs no sign-off from the FHFA. HousingWire, citing unnamed sources, also reports an alternative annual subscription based on prior-year volume, and a date of 31 January for lenders to choose a path. Tonight FICO says that "several deals" have been announced and that it is working with others on programmes for calendar 2027. It does not name them, and neither did the call, so we do not know whether Rocket or UWM, which together are about 95% of Vantage volume so far, are among them. Lansing said that many lenders are in talks, and would not put a number on uptake.

The case for broken. There is no headline rise for the first time since 2018. FICO had to give something away and ask for something back, and a true monopolist does neither. The rival is at a tenth of the price, committed through 2028. Every future rise now has a political cost, and Mr Pulte's reaction to a flat price was to say that more action was coming. JPMorgan now calls it closer to a mid-single-digit revenue grower.

The case for paused. FICO did not cut the $10. It held a premium of ten to one, and early checks say most lenders will pay it. A business with no pricing power does not hold ten to one. About 94% of Fannie's third-quarter loans were still delivered on Classic. Mr Pulte holds an office, not a law. And if loans priced on Vantage turn out worse, which Milliman, AEI and Arch all warn about in different ways, the argument could swing back.

Our reading. "Paused" is too kind and "broken" is premature. The annual, unilateral price lever in conforming mortgages is broken, and a conditional toll survives. Three distinctions matter.

Holding a price and raising it are different powers. FICO has shown the first and lost the second, in the channel that produces about two-fifths of its revenue.

Today's revenue rests on a strange foundation. Lenders pay FICO partly to find out whether another number is higher. Lansing calls this gaming and calls lender choice bad policy, and he is right that it benefits from the behaviour he condemns: to game the two scores you must buy both. His estimate that Vantage would reach somewhere in the 20s as a share was tied to that gaming, and he gave it before Rocket's default and the single grid.

The test is pull share, not delivered share. FICO is paid on the pull. The 6% of Fannie deliveries is the comforting number and the 37% of DU applications is the frightening one, and neither measures the money. The figure to watch is the 200-odd lenders who, on Equifax's account, use Vantage alone. Each of them is a lender who has found that it can stop buying FICO. Bi-merge would be the other mechanical blow. If Fannie and Freddie moved from three bureau reports to two, FICO would lose a third of its scores per borrower on those loans, on our reading. The order was reported as planned in early October. We can find no order as of today, and when asked on the call, Lansing had nothing to report.

The rival's price. In October 2025 the bureaus priced VantageScore at $4 to $4.50 and offered it free with FICO. By 9 March 2026 they had all cut it to about a dollar. By September TransUnion had fixed $0.99 to December 2028, and Equifax $1 through 2028. A firm that cuts its price that fast does not expect to make much money on the product. Equifax says that full conversion from FICO to Vantage would add only about $35m of margin. So their motive looks less like a new profit pool and more like removing FICO's threat to go around them, and escaping the dilution of its royalties. That is why we expect them to keep at it whatever Vantage itself earns.

Who is moving. Rocket made Vantage its default on eligible direct-to-consumer GSE and VA loans from the fourth quarter, keeping FICO on FHA, jumbo and others. UWM said about a quarter of its borrowers do better on Vantage. Pennymac has rolled it out across all channels. Rocket and UWM were about 95% of Vantage volume in September, on KBW's figures, at penetration of 34% and 25%. The whole shift so far rests on two very large companies that had the systems to do it. The test is whether the next hundred lenders follow.

Both sides' evidence needs care. Fannie's 6% is the cleanest number we have, and it is from before the grid. FICO's strongest evidence for 10T comes from Milliman under a consulting engagement with FICO. Vantage's strongest adoption figures come from itself and its owners. On predictive power, AEI's researchers find the two broadly similar. We would not buy or sell this stock on a claim that one score predicts default much better than the other.

The consumer brand is still strong. People know their FICO score. But lender choice moves the decision further from borrowers, who care what rate they get and not which number produced it.

Credit where due. Between January 2012 and the peak Lansing turned $36 into $2,382. He found the pricing power that thirty years of management had left on the table, he retired about two-fifths of the shares, and he did it with almost no capital. In management's defence on the last two years, the threat came from a change in Washington which nobody could have forecast in detail.

But the summer of 2026 is hard to defend. FICO bought back $3.05bn of stock in fiscal 2026, about three times its free cash flow. The biggest single piece came on 5 June, when it borrowed $1.5bn on a two-year term loan and paid it all to Wells Fargo for an accelerated repurchase, at a stock price of about $1,137. FHFA had started the Vantage rollout six weeks earlier, and the bureaus had cut to $1 in March. The danger was on the table. Fifty-five days later the stock fell 17% in a day, and 117 days later it closed at $592. The morning after that, management committed to cutting 15% of positions.

To be fair, the 2.28m shares delivered by June, at about $1,194, were bought by a board acting on a view of value and not a formula, and a low price during the term of the repurchase delivers more shares at settlement. Neither the release nor the call gives the final figure, though Lansing put the year at 2.514m shares for $3.03bn. But the charge against it is not bad timing. A company borrowed two-year money, against negative equity, to buy stock, at the moment when the risk to its franchise became visible. Buffett's test for a buyback is that the shares must be bought below intrinsic value, conservatively worked out. Our central value today is less than half of what FICO paid. That paper loss of about $1.1bn is more than a full year of free cash flow.

The debt. Total debt is $5.26bn, against $3.06bn a year ago, at an average rate of 5.64%, with $269m of cash and marketable investments and $338m due within twelve months. We put net debt at about 3.7 times EBITDA after stock pay. With stock pay added back, as covenants commonly do, it is about 3.3 times, in line with the 3.29× Lansing gave on the call. The covenant limit falls from 4.5× to 4.0× on 31 December 2026 and 3.5× a year later. We cannot compute the covenant ratio because its definition is not in the 10-Q, but on these figures there is room. About $1.76bn falls due within five weeks in mid-2028: the balance of the term loan and $900m of 4% notes, which would refinance at something more like 6%. With about $1bn of free cash flow, this is manageable. It does mean that the buyback, the engine of per-share growth for fourteen years, must stay small for a while. Stress it. If mortgage revenue fell 30%, EBITDA would lose about $280m. With stock pay added back, today's net debt would then be about 4.0×, through the 2027 limit; with about $1bn paid down by that date, it would be about 3.2×. So FICO can survive that outcome, provided it stops buying stock. It has no room to do both. That is why one remark on the call matters. Lansing described the pause as lasting a quarter or two and told holders to expect FICO back in the market using its free cash flow, while the CFO would not set a timeline. The same dollar cannot both repay debt and retire shares. If mortgage revenue holds, that is a fair choice. If it does not, it is the summer's mistake on a smaller scale.

Succession. The CEO is 67, personally runs the segment under attack, has no named deputy for it, and has a retention arrangement that ends in June 2028, the month that $900m of notes fall due. No successor has been reported. In a calm year that would be a footnote. It is more than that when the job is a lender-by-lender negotiation with an industry that feels it has been overcharged, conducted under the eye of a hostile regulator.

Integrity. We see no sign of anything improper. FICO's statements on price have been carefully worded, and in places too clever: "no increase in per score fees for lenders" sat next to mortgage revenue up 97%. Each is accurate. Taken together they explain why the company has so few friends in Washington. We take a little off the score for that, and more for the buyback. The 5 is for capital allocation at the worst moment, not for a career.

This is about as good as a ten-year record gets. Revenue has nearly tripled, operating margin has gone from 19% to 50%, and EPS has grown more than tenfold, about 27% a year. Notice how much of the recent acceleration is one lever. From 2023 to 2026, Scores added $878m of revenue and Software $143m. Without the mortgage price, we think this would be a pleasant 8–10% grower. The past three years of growth cannot be extrapolated, because the thing that drove them has stopped.

The balance sheet deserves a word. Negative equity of $3.87bn is not a hole in the usual sense. It is the record of $9.5bn paid out for stock in a business that needs almost no assets. Total assets are just $2.08bn, of which $792m is goodwill. It does mean that there is no cushion of any kind between the lenders and the franchise, and that the value of the shares is entirely the value of future tolls.

Dividend. FICO stopped paying in May 2017, so our six dividend tests do not apply. All capital return has gone through buybacks, and for the moment that channel is closed by the debt. A shareholder gets no cash while waiting, which matters for anyone used to being paid to be patient.

Why ~$40. We reach about the same place from two directions. From cash: free cash flow less stock pay gives about $850m. From earnings: net income of $835m, plus the settlement and restructuring, less the gain, gives about $870m after tax. A full year of the September quarter's interest takes off about $70m, and a full year at $10, against one quarter at the old price, adds about as much back. That points to $40–$42 on today's share count, and we use about $40. It is before any 2027 price concession.

The enterprise model. We start from a run-rate of about $1.05bn of mortgage revenue (not the June quarter annualised, which is the seasonal peak), $0.69bn of other Scores, and $0.94bn of Software. We take off about $0.40bn a year for corporate costs, stock pay and amortisation after the staff cut, and tax at 26%, the rate the CFO gave for fiscal 2027. We discount operating profit after tax to 2031 at 10%, add a terminal multiple, and subtract net debt of about $5.06bn including the settlement. The CFO's remark about reinvesting the savings makes our 30% Software margin a little less safe.

The central value moves to about $600 at a 9% discount rate and $535 at 11%; to about $495 at an exit multiple of 15× and $670 at 20×. If mortgage revenue merely stays flat, it is about $625; if it falls 6% a year, about $515. What is priced in? At $701, on the central assumptions for everything else, mortgage revenue must grow about 3.5% a year if the business then sells for 17 times profit, or about 7% a year at 15 times. So at this price you need the toll to rise, or the multiple to stay high, and the company has just told us that it cannot yet say which way the price will go.

Why such a wide range? Because of the debt. Net debt takes about 30% of enterprise value, and it stays fixed while the rest moves. Each 5% change in the value of the business is about 7% on the shares. The low case is $190 and not $400 for that reason. It is also why we would not use the fall from the top as a guide. A stock that has lost 62% can lose half again, and in the low case it does.

Multiples on the consensus. On estimates made before tonight, the shares trade on about 15× fiscal 2027 EPS ($45.76) and 14× fiscal 2028 ($50.59). Those will move. Analysts' mean target before the results was about $1,025, in a range of $700 to $1,675, and it has been falling fast.

A comparison. Our X-Ray of Booking Holdings, written the same day, has that stock at about 17 times owner earnings. FICO sits on nearly the same multiple. Booking has net debt of 0.4 times EBITDA, and needs 4% growth to pay 10%, with no agency able to rewrite its price. FICO has 3.7 times, and a regulator who has already done so. The same multiple does not make them the same value. Our figure for FICO is below the price, and for Booking above it.

Why $425? It is 25% below our central value and under 11 times owner earnings. For a business we rate 2 out of 5 on what can be known, the usual rule would ask for more, and we would not argue with anyone who waited for less. But at $425 the market would be pricing something close to a steady decline in mortgage revenue, with the Software and non-mortgage lines thrown in cheaply, and the buyer would make about 2.4 times the money in the high case and lose a little over half in the low one. The closing low so far is $592.47, so the market has not offered that yet.