Season 3 · Episode 4

Mutual Funds Cut Private Software 20%. Buyout Marks Moved 8%.

Same sector, same quarter, two different numbers, because one holder has to publish a mark and the other does not. Underneath both sits a loan written against annual recurring revenue, at a moment when annual recurring revenue has stopped describing the business. Four borrowers have already switched their covenant off it. That is the move, and almost nobody has made it.

The Agentic Commercial Model Newsletter · Fessal Rahman · November 20, 2026 · 12 min read

From FR Advisory.

The author spent twenty years running commercial organisations inside software and private equity portfolios. He no longer works in the technology industry, holds no position in any company named in this series, has no clients in it, and is selling nothing to anyone in it.

Start with the seller, because the seller gets the instruction first and the explanation never.

A commercial director in a sponsor-backed software business spends the autumn building the case for the crossing. The product team has shipped agents. The customers are asking for a meter. The economics work on a three year view. The deck is good.

It dies in a meeting she is not invited to, and the reason given to her afterwards is that now is not the right time.

The real reason is a document. Somewhere in the capital structure there is a loan, and that loan is sized, tested and covenanted against annual recurring revenue. Convert committed seat spend to consumption and the recurring revenue line falls before the consumption line is large enough to replace it. That dip is a covenant question, a mark question and a refinancing question, in that order, and none of them are hers.

She has been told to protect ARR. What she has actually been told is not to cross.

This is the part of the transition nobody writes about, because it has no product in it. The commercial model is not only being held back by instruments that cannot see the crossing, which is where Episode Five goes next. It is being held back by instruments that have money lent against them.

The gap, measured

In the first quarter of 2026, two sets of holders valued private software and arrived at very different answers.

Mutual funds, which must disclose their marks, cut nearly fifty private software companies by an average of 20%, based on public filings across hundreds of portfolios. The steepest were severe: Outreach down 51.4%, DataRobot down 50.8%. Epic Games down 22%, Databricks down an average of 16%, Canva down 15%.

Over the same quarter, Bain found software valuations in private equity portfolios declined roughly 8%. In the United States specifically, 8.9%. In Europe, 4.2%.

Twenty against eight. Not the same companies, and that distinction matters: the mutual fund sample is venture and growth-stage private software, the buyout sample is sponsor-controlled software. They are different books. But they are the same asset class absorbing the same shock in the same ninety days, and the holder with a disclosure obligation moved more than twice as far as the holder without one.

The public comparison is harsher still. The S&P 500 software index fell 27% between early January and late March 2026.

So: public software down 27%, disclosed private marks down 20%, sponsor marks down 8%.

Why marks move slowly, and why that is not a scandal

The tempting conclusion is that private marks are fiction. That is too easy, and the evidence does not support it cleanly.

Marks lag for structural reasons, not conspiratorial ones. A quarterly valuation uses a comparable set drawn from a period that has already closed. Valuation policies smooth deliberately, because a mark that tracked daily public volatility would be useless for the ten year fund it sits inside. A sponsor with no obligation to sell has no forced-seller discount embedded in its carrying value, and that is legitimate, because it genuinely is not a forced seller.

And here is the fact that complicates the cynical reading, from Bain's own midyear work: more than 75% of buyout assets are still exiting above their next-to-final quarterly mark. When sponsors do sell, they sell above the carrying value. On its face that says marks are conservative rather than inflated.

Read it alongside the other finding and it says something sharper. Exits slowed through the first half. Deals clear mainly for high quality assets, often only where sellers accept lower valuations or use a continuation vehicle. Tech deal value fell about 70% from the fourth quarter of 2025 to the first of 2026.

So the assets that exit above their mark are the assets that could exit. The ones that cannot clear their mark do not become down-rounds. They become holds. A statistic measuring only the companies that sold tells you about the companies that sold.

That is not fraud. It is selection, and it is exactly what you would expect from a system where nobody is obliged to transact. But it means the 75% figure cannot be used as evidence that the whole book is marked correctly, which is roughly how it gets used.

Dan Niles, of Niles Investment Management, put the incentive plainly in July: "You're not going to hear any management team come out and say, yeah, we're dead." Aram Green at ClearBridge was blunter about the position sponsors are in: "They know there's no appetite and there's no clear way out."

Against that, Michael Brandmeyer, who runs Goldman Sachs' External Investing Group, called the reality of AI disruption "much more nuanced", and both Orlando Bravo and Robert Smith have defended their software books publicly. They are not obviously wrong. Thoma Bravo also booked a loss of more than five billion dollars on Medallia, which the firm said was unrelated to AI. Large losses are being taken. They are simply being taken one at a time, in a sequence the holder controls.

The loan is written on the metric that stopped working

This is where the capital story collides with the rest of this season, and it is the part almost nobody is connecting.

A large share of sponsor-backed software is financed with recurring revenue loans. These lend against annual recurring revenue rather than earnings, on the reasoning that a subscription book is predictable enough to underwrite directly. For a decade that reasoning was sound. ARR was contractual, renewals were the event, and the metric was stable enough to carry a covenant.

ARR has stopped describing businesses that are crossing to consumption, and Episode Five sets that out company by company from the filings. Not because anyone is lying, but because the instrument was built to measure committed subscription value and a meter does not produce committed subscription value. HubSpot's retention fell when customers swapped seats for credits. Salesforce reports a consumption product as recurring revenue and widened the definition of that number in the same quarter it printed the growth.

Now put a loan on that.

It is worth being precise about how these loans work, because the mechanism is where the damage happens and most commercial leaders have never seen the inside of one.

A recurring revenue loan sizes the facility as a multiple of ARR rather than a multiple of earnings, on the logic that a subscription book is annuity-like enough to service debt directly. The facility carries an ARR covenant, tested quarterly, and usually a liquidity covenant alongside it. Critically, most of these structures contain a step-down: after an agreed period, or on an agreed trigger, the test converts from an ARR basis to a conventional leverage basis measured on earnings. The borrower is expected to grow into profitability and graduate onto normal terms.

That design has a specific blind spot. It assumes the only two states are growing recurring revenue and shrinking recurring revenue. It has no state for recurring revenue falling because the customer moved to a better commercial model. A conversion that increases the lifetime value of an account and extends its life reads, inside the covenant, exactly like a customer leaving. The document cannot tell the difference between a cancellation and an upgrade to a meter, because the number it tests was never built to.

So the borrower faces a choice the loan never anticipated: run the crossing and risk tripping a test on an improvement, or hold the seats and stay compliant while the business ages.

KBRA, which rates securitisations of these loans, publishes a dashboard on the underlying collateral. Across 23 rated recurring revenue loan transactions covering 102 obligors, on loan tapes dated March 2026: average annual recurring revenue of $177.6m, debt to ARR of 1.7 times, average loan to value of 29.7%.

And this, which is the sentence in the whole episode:

Four borrowers switched from ARR-based covenants to EBITDA-based covenants in the quarter, the same number as the period before.

Borrowers are moving their covenants off the metric. Not many. Four, then four. But the direction is the tell, because a borrower does not renegotiate a covenant basis for fun. It costs fees, it costs concessions, and it invites a lender to reopen terms. You do it when you can see that the number the covenant is written on is about to move for reasons that have nothing to do with your business getting worse.

The rest of the collateral supports the urgency. Aggregate ARR across that population fell 0.9% in the quarter and 0.7% year on year. Balance sheet cash averaged $33.2m, down 17.4% on the year and well below the $47.9m historical average. The liquidity cushion fell 13.8% in the quarter. Around 34.5% of these loans carry capitalised interest, averaging 5.73% of principal, which means a third of the book is partly paying its interest with more debt. KBRA reported no defaults.

So: recurring revenue shrinking slightly, cash down sharply, a third of the book capitalising some interest, and a slow migration away from ARR covenants. That is not a crisis. It is a system quietly repricing the assumption it was built on.

For scale, private credit is roughly a three trillion dollar market and software has been running at twenty to twenty-five per cent of private credit deal flow on 9fin's figures. UBS sketched an aggressive disruption scenario in which private credit default rates could reach 13% in 2026, against a 4% stress case for high yield. That is a scenario, not a forecast, and it should be read as one. Michael Arougheti, who runs Ares, has said his firm's software exposure is about 6% of assets and 9% of its private credit portfolio, which is a long way from systemic.

The point is not that private credit is about to break. It is that several hundred billion dollars of software debt is covenanted on a number that this season has spent five episodes showing cannot see the thing the industry is doing.

What this does to the sales floor

Everything above arrives on the commercial floor as a single instruction: protect ARR.

That instruction is rational given the capital structure and catastrophic given the market. It means hold the seats. It means price the agent as a subscription add-on rather than a meter, because an add-on preserves contracted value and a meter does not. It means the account executive who converts committed spend into consumption has damaged the covenant headroom and will be compensated accordingly, which is to say badly.

Three things follow, and they are the mechanics of the death spiral this season has been circling.

The comp plan enforces the capital structure. Not the strategy, the capital structure. If the plan pays on contracted ACV and the covenant is tested on ARR, then every incentive in the commercial organisation is pointed at preserving a number the buyer has stopped wanting to pay. Sellers are not resisting the transition. They are responding correctly to the plan in front of them.

The customer leaves anyway. The buyer asking for a meter does not stop asking because your lender prefers subscriptions. They buy the meter from a competitor without the debt, or from a vendor that never had seats to lose. The covenant does not protect the revenue. It only delays the recognition.

The exit gets harder the longer you wait. This is the part sponsors underweight. An asset that crossed two years ago is messy for four quarters and legible afterwards. An asset that has not crossed is clean on the current scoreboard and unsellable to any buyer who has read a HubSpot filing, because the buyer is underwriting the crossing you have not done and will price your delay into their bid. A majority of buyout assets, by count and by value, were bought in 2021 or earlier. Every quarter of delay adds to a holding period Bain already describes as an implied capital cycle of roughly seven years, with distributions as a share of net asset value at record lows for about four years.

You are not protecting the asset. You are ageing it.

And the destination for an asset that cannot clear its mark is increasingly not a sale at all. PwC's midyear work describes continuation vehicles and sponsor-led secondaries as having become the primary liquidity mechanism, with limited partners increasingly questioning the conflicts in them. For the operator that is not a financing footnote. It is a new owner with the same thesis, a reset clock, a fresh set of value creation targets written against the model you were not allowed to change, and in many cases the same board asking why growth has slowed. The covenant that stopped the crossing in year five will still be there in year seven, attached to a vehicle that exists precisely because nobody would buy the asset as it stands.

Underneath all of it sit the layers this series has been building, and in this episode they run in an unfamiliar order. Revenue architecture comes first, because the capital structure is now the binding constraint on commercial design rather than a consequence of it. Compensation comes second, because the plan is where the covenant actually reaches the customer. Pricing and packaging third, since the shape of the meter determines how deep and how long the trough is. Then metering and entitlement, because the lender conversation is unwinnable without a meter whose numbers survive scrutiny. Organisation design comes last here, and it matters mainly in one respect: somebody has to own the conversation between the commercial plan and the credit document, and in most sponsor-backed businesses that person does not currently exist.

The move, and almost nobody has made it

Four borrowers a quarter is the entire visible population of companies doing the obvious thing. Here is what the obvious thing is.

Renegotiate the covenant basis before you cross, not during. A lender presented with a credible conversion plan, at a point when the credit is performing and the headroom is intact, has a commercial conversation available to it. The same lender presented with a covenant breach caused by a conversion already underway has only a workout conversation. The sequencing is the whole game. The four borrowers who moved to EBITDA covenants in the quarter did it from a position of strength, which is the only position from which it works.

Bring the lender a crossing disclosure set before they ask for it. Converted spend, retention on two bases, consumption volume and revenue separated, realised rate. Episode Five specifies all six in full. A lender's objection to the crossing is not ideological, it is that the crossing is unmeasurable in the terms of the loan. Measure it in terms the loan can use and you have changed the conversation from trust me to here is the data. Every one of those numbers exists in your billing system today.

Make the sponsor choose explicitly. The conversation that does not happen is the one where somebody says out loud: we are choosing a lower mark in eighteen months in exchange for an asset that is sellable in thirty-six, or we are choosing the opposite. Sponsors are good at that conversation when it is framed as a trade. They are terrible at it when it arrives as a comp plan nobody authored on purpose.

Fix the plan in the same quarter you fix the covenant. A covenant moved to EBITDA while the plan still pays on contracted ACV changes nothing on the floor. The seller does what the plan pays for. Converted spend has to become a carried, paid component the quarter the covenant allows it, or the freedom you negotiated goes unused.

The first ninety days

Days one to thirty. Find out what your covenant is actually written on. Not what finance says, the document. Which definition of recurring revenue, tested how often, with what headroom, and what happens to that definition when a customer converts a committed seat contract into a credit balance. A surprising number of commercial leaders in sponsor-backed businesses have never read this and are being governed by it daily.

Days thirty to sixty. Model the dip. Produce the conversion curve for a representative cohort: committed spend out, consumption revenue in, quarter by quarter, with the trough stated plainly. If the trough breaches, you have a financing problem, not a strategy problem, and it has to be solved in that order. If it does not breach, you have just removed the objection that has been killing the proposal.

Days sixty to ninety. Take it to the lender with the sponsor in the room. Covenant basis, the disclosure set that will evidence it, and the comp change that will make it happen. All three, together. Any one of them alone fails.

The position

The commercial model is not being held back by technology, and at this point it is barely being held back by belief. It is being held back by documents. A covenant written on annual recurring revenue, a carrying value computed from a comparable set that has not caught up, and a compensation plan that quietly enforces both.

In the first quarter of this year the holders who had to publish a number published twenty per cent. The holders who did not published eight. The public market published twenty-seven. Those three numbers describe the same sector, and the distance between them is the amount of recognition still to come.

Meanwhile the collateral tells its own story. Recurring revenue across a rated sample fell slightly. Cash fell sharply. A third of the book is capitalising some of its interest. And four borrowers a quarter, quietly, are moving their covenants off the metric before the metric moves under them.

Four is not a trend. It is a tell.

The sponsors who cross will take a lower mark on the way through and will own an asset somebody wants in 2028. The sponsors who protect the mark will own a clean scoreboard attached to a business the buyer has already learned to discount. Both will describe their choice as discipline.

Only one of them will get paid for it.

Next: Season 3, Episode 5. Salesforce sells it by the action and reports it as recurring revenue. The six numbers a crossing has to publish.

The Agentic Commercial Model. agenticcommercialmodel.com

Sources

Next: Episode 5 →Subscribe on LinkedIn →