On 23 May 2024 Workday reported a quarter most companies would frame. Revenue up 18% to $1.99 billion. Subscription revenue up 19%. About 10,500 customers. Then the chief executive, Carl Eschenbach, told analysts the company was seeing customers committing to lower headcount levels on renewals than it had expected. The chief financial officer, Zane Rowe, said customer headcount growth had slowed, and that this, not lost deals, was what had hurt. Workday cut its subscription revenue guidance for the year to $7.7 billion to $7.725 billion, from $7.725 billion to $7.775 billion. The shares fell 11% after hours.
Nothing in the revenue line had moved yet. The seats had.
Here is the thesis. Every software loan, every software valuation and most software diligence is built on a dollar number: recurring revenue, and the share of it that renews. Dollars are the last thing to move when a customer needs fewer people. Seats move first, and the price increase that arrives at the same renewal hides them. Lenders and owners of software assets should underwrite and monitor seats separately from price, because by the time the dollar retention number breaks, the loan is already a restructuring.
The arithmetic that hides the seats
Start with one renewal. A customer pays for 1,000 seats. At renewal it needs 940, because it has stopped hiring for a role, or automated part of one. The vendor raises the price per seat by 6%. The new contract is worth 940 times 1.06, which is 99.6% of the old one.
On a net revenue retention slide that is a customer who stayed and paid roughly the same. In the seat table it is a customer who cut 6% of its licences in one cycle. If the same customer adds a module, the dollar number goes above 100% and the slide calls it expansion.
Now add the two features every enterprise contract carries. The first is the term. Enterprise software is commonly sold on annual or multi year agreements, so a customer who needs fewer seats in month four keeps paying for the old count until the renewal date. The second is the floor. Contracts set minimum commitments that a customer cannot go under mid term. Eschenbach said exactly this on Workday's call on 25 November 2025: customers can bring headcount down, and floors and minimums give the company some protection.
Protection is the right word for the vendor. For the lender it is a delay. A floor means the seat loss happened and has not been invoiced yet. On three year terms only about a third of the book reprices in any one year, so the dollar retention number becomes a moving average of decisions customers made up to three years earlier.
So the order of events is always the same. The customer's need falls. The seat count falls at the next renewal. The price rises at the same renewal and covers most of it. Revenue keeps growing because new logos and add ons are still landing. Dollar retention dips a point or two. And only when the price lever is exhausted, or a competitor makes the product free, does the revenue line follow the seats down.
You can see what happens when there is no price lever to pull. Chegg sells to students, one subscription each, and cannot raise the price on someone who has stopped coming. In the second quarter of 2025 its subscribers fell 40% to 2.6 million and its subscription revenue fell 39%. Units and dollars moved together, in the same quarter, because nothing sat between them. Chegg cut about 45% of its staff in October 2025. Enterprise software has a contract, a floor and a price increase sitting between the units and the dollars, and that buffer is what makes it look safer than it is.
Why lenders are looking at the wrong line
This matters more now because of how much of the software industry is financed by people who only see the dollar line.
The Bank for International Settlements put outstanding private credit loans to software as a service companies at over $500 billion at the end of 2025, up from almost $8 billion in 2015. That is 19% of all direct loans. A third of private credit funds have lent to the sector. In a July 2026 bulletin, BIS economists Fernando Avalos, Giulio Cornelli and Egemen Eren put business development company lending to software at around $115 billion: about a fifth of everything those vehicles lend, and over 80% of their technology books.
Their two most important findings sit side by side. Revenue uncertainty from generative AI has not affected these loans yet. And neither the lenders nor their equity investors have priced software exposure differently. Spreads have narrowed, which leaves less buffer to absorb losses.
Both findings are what my arithmetic predicts. Of course it has not shown up in the loans yet. The loans are measured on the number that moves last.
A large part of this lending was written as recurring revenue loans. The borrower is a growing software company with little or no operating profit, so the lender sizes the debt against annual recurring revenue instead of earnings, with financial covenants tested on that revenue until the company is expected to make money, at which point they convert to normal cash flow covenants. PitchBook called the structure a hallmark of the easy credit years. It has not gone away. Permira's merger of Highspot and Seismic this year is financed with more than $600 million of debt priced at 475 basis points over SOFR, and PitchBook reported it carries initial recurring revenue covenants that convert to cash flow covenants later.
Read that structure with the seat arithmetic in mind. The covenant tests a number that a 6% price increase can hold flat while 6% of the seats walk out. It tests the dollar result of a renewal and says nothing about its composition. A borrower can pass every quarterly test for three years while its installed base shrinks underneath it.
The market has already noticed something is wrong, even if the loans have not. BIS measured software company stocks down almost 30% between October 2025 and February 2026, and business development companies with high software exposure underperforming those with low exposure by about 5 percentage points over the same period. Software's share of private equity backed direct lending volume fell to 15% so far this year from 22% last year, per LCD estimates reported by PitchBook. Lenders are writing fewer new software loans. They are still marking the old ones on revenue.
Three cases
Pluralsight: the loan that was marked on dollars
Vista Equity Partners took Pluralsight, an online technology training company sold largely to enterprises by the seat, private in April 2021 at a valuation of about $3.5 billion. The debt came from private credit, not banks: a recurring revenue loan of about $1.2 billion plus a revolver, per the Financial Times as cited by the law firm Goodwin, with Reuters putting the total above $1.5 billion. The lenders were a who's who of the industry: Blue Owl, Ares, Golub, BlackRock, Oaktree, Goldman Sachs asset management and Benefit Street Partners.
The business it lent against was a seat business. Its customers were technology teams, and the people in those teams were the seats. Through 2023 those customers were cutting technology staff. Competitors made learning products free in the first quarter of 2024. By then the new chief executive was forecasting a customer renewal rate for 2024 in the low 70s, per Reuters.
Look at the order in which the numbers moved. Vista began reporting revenue declines to its own investors in the third quarter of 2023. It had already asked lenders to loosen covenants in the first quarter of 2023 and put in $75 million of fresh equity. Fund documents seen by Reuters showed Vista marking its equity at $2.3 billion in the first quarter of 2023, $1.5 billion by the fourth quarter, and zero by the first quarter of 2024.
The lenders' marks moved last. Legal and General's investment research team wrote that the credit risk was visible from 2023, but the lenders only marked the loan down to 45% to 50% of its value in the second quarter of 2024. In August 2024 they took 100% of the company, wrote $1.3 billion of debt off and reportedly put in $250 million of new money.
The sequence is the whole argument. Customers' headcount went first. Seats followed at renewal. Equity holders, who look at the forward book, saw it next. The loan, tested and valued on recurring revenue, saw it last: a quarter after the owner had written its equity to zero, and more than a year after the owner started marking it down.
Workday: a vendor saying the quiet part
Workday is not a distressed borrower. It is a profitable public company, and that is why it is useful. It prices most of its core products by the employee, so its seat count is its customers' headcount, and it has been unusually open about what happens when that headcount stops growing.
In May 2024, as above, it told the market that renewals were coming in at lower headcount than expected and that it expected the pattern to persist. Rowe described the growth that remained as coming primarily from product add ons rather than customer headcount. That is the composition shift in one sentence: fewer seats, more products per seat.
On 25 November 2025 the company reported 12.6% revenue growth to $2.4 billion for the quarter to 31 October. Eschenbach said customer headcount was still growing modestly, conceded under questioning that some customers were laying people off, and then said the line that should be pinned above every software credit committee: Workday was focused not just on seats but on revenue per seat. The shares fell about 9% the next day.
There is nothing wrong with that strategy for Workday. It is exactly what a good management team should do. But it describes a business that has stopped expecting the seat count to carry growth and has moved the weight to price and attach rate. If you lent against that revenue, or own a smaller company that sells the same way, you are now underwriting a pricing strategy, not an installed base. Those are different risks with different failure modes, and neither shows up in a recurring revenue total.
Salesforce: the price lever, pulled in public
Salesforce raised list prices on the Enterprise and Unlimited editions of Sales Cloud, Service Cloud, Field Service and some Industries clouds by 6%, effective 1 August 2025. The previous increase was 9% in 2023. At the same time it retired standalone AI add ons and moved customers to bundled Agentforce plans priced from $125 to $550 per user per month, and raised Slack Business+ by 20% to $15 per user per month.
In May 2025 it launched Flex Credits, which let customers convert unused licences into AI capacity. That product exists because customers hold licences they are not using. On the December 2025 earnings call the chief revenue officer, Miguel Milano, said a lot of customers still want seat based pricing for its predictability, said prices on the core clouds are rising, and described unused licences being converted to credits, to a new enterprise licence agreement, or back to seats.
I am not saying Salesforce has a seat problem. It has the most pricing power in enterprise software and is using it. I am saying it has published, in plain view, the playbook every seat priced company in a private equity portfolio will now run: raise price on the seats that remain, bundle the new thing into the old licence, and give customers a way to convert idle seats into something that still counts as revenue. Every step of that playbook holds the dollar line up. Not one of them tells you whether the customer needs as many people using the product as it did last year.
When the largest seat vendor in the world is designing products to absorb idle seats, a lender to a $40 million revenue HR or collaboration software company should assume its borrower's customers have idle seats too.
The method
I have written software since 1995, and for most of that time the licence count was the dullest table in the building. It sat in the billing system, it went up every year, and nobody senior asked for it, because revenue already contained it.
That assumption is the thing that broke. When I read a software asset now, the first table I want is not the recurring revenue bridge. It is seats by customer by month, licensed against active, for the last 36 months, joined to the renewal date and the contracted floor. From that table you can split dollar retention into three numbers: what came from price, what came from more products per customer, and what came from people. The third number is the one that tells you whether the customer still needs the product at the same scale. In a seat priced business it is the closest thing you have to a leading indicator, and it is the one that never appears in a lender's compliance certificate.
The second thing I do is line the seat movement up against the renewal calendar. If seats are falling at every renewal and the price is rising at every renewal, then the current dollar retention is borrowing from the next price increase. That works until the price increase stops landing, which is usually when a competitor or a model makes part of the product free.
None of this needs new data. Every software company already holds its seat counts, because it has to invoice them. The problem is that nobody downstream of the billing system asks for them in a form that separates seats from dollars.
Where this breaks
Four honest limits.
First, not all software is sold by the seat. Usage priced infrastructure, payments software and anything billed on transactions or data volume have a different unit, and the seat table is meaningless for them. FTI Consulting's September 2026 review of 133 public enterprise software companies gave security software an average AI risk score of 1.33 and infrastructure software 1.49, against 2.12 for customer relationship management. The argument here applies most to seat priced, headcount linked software: CRM, HR, collaboration, learning, service desks. It does not apply evenly across a software book.
Second, my cases mix causes. Pluralsight's customers cut technology staff in a rate shock, and its competitors went free. Workday's 2024 renewals reflected a hiring slowdown, not agents doing the work. I am not claiming AI caused any of them. I am claiming the instrument missed all of them, for the same reason, and AI is a new cause of exactly the same seat movement.
Third, price increases can be real. If the product does more, customers will pay more per seat and the dollar line is telling the truth. Salesforce says customers get several times more value from AI bundles. Maybe so. The way to test it is the seat table: if price rises and seats hold, the value is real. If price rises and seats fall, the customer is paying more per person for fewer people, and that ends.
Fourth, borrowers will resist. Seat data by customer is commercially sensitive, and a sponsor with a strong hand will refuse a new reporting line. That is a negotiating problem, not an analytical one, and the refusal is itself information. FTI already lists seat count reporting triggers among the covenant tools lenders should consider. It is not an exotic ask.
The fair summary of the other side is the BIS finding: nothing has shown up in these loans yet. That is true. It was also true of Pluralsight's loan in 2023.
What I would do on Monday
If you lend to software, ask every seat priced borrower in the book for licensed and active seats by customer by month for 36 months. Start with the ones whose recurring revenue covenants have not yet converted to cash flow tests.
Split dollar retention into price, product attach and seats for each of them. Put the seat line in the credit memo next to the revenue line. If seat retention is below 95% while dollar retention is above 100%, flag it.
Build a flat price case. Hold price per seat at today's level and roll the seat trend forward through the renewal calendar. That is the revenue the business has if the next increase does not land.
Write a seat count reporting line into every new software financing and every amend and extend. It costs the borrower a query against its billing system.
If you own the company rather than lend to it, take the price increases out of the value creation plan's organic growth line and show them separately. A plan that grows by pricing seats that are shrinking is a plan with a fixed number of years left in it.
The sentence on the call
Go back to Workday on 23 May 2024. Revenue was up 18%. The company could have said nothing about seats and nobody would have asked. Instead its chief executive told the market that customers were renewing at lower headcount than expected, and its share price fell 11% on a quarter of strong growth.
That was a public company with analysts on the line asking direct questions. A private borrower has a compliance certificate and a quarterly revenue number. Nobody on that call asks about seats, so nobody hears the sentence.
Ask for the seat table before the revenue tells you.
I build software for private equity owners and spend much of my week reading how software businesses actually earn their revenue, not how the revenue line looks in the data room. If you hold a software loan or a software company and have never seen its seats split from its price, show me the seat table.

