← Back
Thesis. When a software company bills for the human labour its own product helps customers remove, AI creates a problem that appears in none of the usual numbers. Customer productivity rises while vendor revenue can stall or fall, inside the same contracts, with nobody leaving. Not every software company has this problem. The ones that do rarely find out from a dashboard. And where the old billing unit is already contracting, replacement revenue may not scale quickly enough to avoid a trough, which makes this a capital allocation question rather than only a product question. Founders who quantify it early keep several good options. Founders who defer it are not avoiding a decision. They are allowing arithmetic to make one.
The board met on a Thursday in July and the numbers were good.
Cristina had built her insurance broking platform over sixteen years. 410 brokerages across Australia and New Zealand. $19m of recurring revenue. Gross retention of 93%. The business had been profitable for nine consecutive years and had never taken outside capital, which Cristina regarded, correctly, as an achievement.
Item six on the agenda was AI. Her CTO presented for eleven minutes. Two features had shipped that year: an assistant that drafted submission summaries, and a claims triage tool that sorted incoming notifications by complexity. Both worked. Customers liked them. One brokerage had written to say the triage tool saved its team about a day a week.
The board was pleased. Somebody used the word encouraging. The meeting moved to item seven.
Nobody asked what happens to a brokerage that saves a day a week, every week, across 410 firms, over three years.
Cristina charged per licensed broker.
Eleven months later her renewals manager flagged something small. Across her 40 largest customers, licence counts were down 4% year on year. Not one of those firms had left. Not one had complained. Several had reported record years. They were writing more policies with fewer people, and her invoice was attached to the people.
The same clients. The same contracts. Less revenue.
It would be easy to read Cristina's story as a verdict on seat-based software. It is not, and the evidence against that reading is recent and local.
On 6 August 2026 Atlassian reported quarterly revenue of about US$1.8bn, up 28%, and quarterly net income of US$139m against a loss in the prior-year quarter. The shares rose sharply. On the same day Mike Cannon-Brookes announced his intention to buy up to US$250m of Atlassian shares on market with his own money. This is a company that had been heavily repriced over the preceding year amid fears that AI would make its products less necessary. Seat expansion is still contributing to its growth.
So the question is not whether AI is coming for software. It is narrower and more useful than that.
It is whether a particular company's billing unit is attached to the labour its own product removes.
Where a product makes people more valuable, AI can increase what each seat is worth, and the vendor can charge for it. Where AI allows customers to increase output without increasing headcount, the customer's economic activity grows faster than the vendor's billing unit, and the vendor absorbs the difference at renewal without losing a competitive deal, receiving a complaint, or seeing a warning anywhere in its reporting.
Two businesses can look identical on growth, retention and margin and sit on opposite sides of that line. Cristina's product summarises submissions and triages claims, which is work that brokers used to do by hand. Every improvement she ships makes each broker more productive. Her customers can now write more policies without adding brokers at the rate they once did. Nobody has to lose a job for this to reach her revenue. Headcount growth only has to come loose from the volume of business being written.
The improvement and the erosion are the same event, funded by the vendor.
Her customers' efficiency was arriving on her invoice.
Suppose the diagnosis comes back badly. The remedy is to move revenue onto something that does not shrink when the customer becomes efficient: a platform fee, a transaction charge, a usage line, a premium tier for the automated work, or some combination. Very few established vendors will move to pure outcome pricing, and most should not, because attribution is difficult and enterprise buyers dislike unpredictable invoices. Hybrid is the realistic destination.
The problem is not the destination. It is the journey.
Writing for institutional investors in March 2026, PitchBook describes what it calls the SaS J-Curve: seat revenue declines before replacement revenue scales, and the gap between them is the primary financial risk of the transition. Three things make that gap deeper than a founder expects.
The first is self-cannibalisation. The features that let the customer increase output without adding headcount at the old rate are the vendor's own, shipped at the vendor's expense, attacking the vendor's own billing unit.
The second is the cost of delivering intelligence. Traditional software carried almost no marginal cost per additional user. Software that reasons does. Compute becomes a variable cost of goods sold, and PitchBook's stated threshold is blunt: a gross margin falling below 60% because of AI costs means the model is broken.
The third is the internal machinery. Sales compensation, revenue recognition, forecasting and renewal processes were all built around a unit being retired. For the duration of the crossing, the company navigates by instruments calibrated for the wrong measurement.
If Atlassian ever needs to make such a crossing, it has the balance sheet and platform position to absorb it, and nobody will force a decision on it while it does. A company with $19m of revenue, no outside capital and a sound but unremarkable cash position is in a different position entirely.
That is not a criticism of the company. It is arithmetic.
A trough is not a product problem. It is a funding problem.
Here is where the discomfort begins.
PitchBook's report is written to help investors decide which software assets to hold and which to exit. Among the terms it defines for that purpose is the value trap: a company with strong current cash flows whose revenue is tied to human seats, whose addressable market contracts as its customers become more efficient, and which consequently looks inexpensive against its earnings. The report's guidance on assets fitting that description is to sell them, unless the company is already moving decisively away from seat-based pricing.
Now set that beside the description Cristina would give of her own business. Profitable for nine years. High retention. Long-tenured customers. No debt. A modest multiple.
The facts are identical. Only the classification differs.
That is worth sitting with rather than arguing about. The characteristics Cristina reads as resilience are the characteristics an institutional investor or buyer reads as exposure. Neither is being unreasonable. They are pricing different lengths of time, and only one of them writes the cheque.
A founder may reasonably conclude that investors have overreacted, and she may well be right. It does not change her position.
Across the four quarters to mid-2026, the median enterprise value to revenue multiple of Software Equity Group's index of 106 public software companies fell from 5.7 times to 3.2 times, a decline of more than 40%. Over the same period median revenue in that index reached a record US$1.2bn, median EBITDA rose by close to 70%, and revenue growth improved sequentially for the first time in several quarters. The companies got better. The multiple did not.
In Australia the compression was sharper. Latimer Partners, using S&P Capital IQ data to 11 March 2026, reported small-cap listed software companies in the $50m to $200m market capitalisation band trading at a median enterprise value to revenue multiple of 1.6 times, against a historical range of three to five times, with the ASX All Technology Index down 19% over twelve months.
This matters to a private company for one reason. Those comparables influence the price at which a private company can raise capital to fund the crossing. The window in which the transition is cheapest to make is the window in which the money to make it is dearest to raise.
Meanwhile the buyers stopped debating. Constellation Software has been an acquisition-led business for three decades, so its appetite is not news. The scale of its deployment into a repriced market is. For the quarter ended 31 March 2026 it completed acquisitions for US$697m cash plus US$112m deferred, giving total consideration of US$809m, against US$133m in the same quarter a year earlier. Quarterly deal activity is lumpy, and a single quarter proves little on its own, which is what makes the next line of the same announcement the more telling one: after 31 March the company completed or committed a further US$786m of total consideration, including US$627m in cash. In Australia, Latimer Partners recorded a median control premium of 67% on software takeovers, which shows how far strategic control value can sit above a depressed trading price.
Inside processes, the same repricing shows up as friction. Oliver Wyman reported in April 2026 that buyers and their investment committees are asking materially harder questions about AI exposure, producing more withdrawals from processes and wider gaps between what sellers ask and what buyers offer, with lenders applying the same scrutiny. The models they name as most exposed are seat-based ones whose revenue scales with the customer's headcount.
The founder is waiting for the market to change its mind. The market is transacting on the view it already holds.
The instinct at this point is to commission a strategy. Before doing that, answer three questions. Each can be answered from information the company already holds.
What share of revenue is billed against a unit that shrinks when the customer becomes more efficient? Not an estimate. The contracted number, customer by customer, tracked over 8 quarters.
If the diagnosis is bad, how deep and how long is the crossing on the company's own figures? Model the exposed revenue declining at the rate the last two years have actually produced, add compute at realistic volumes, and find the bottom.
What is available to fund that bottom? Cash, plus debt capacity, plus profit the shareholders are willing to forgo, plus whatever outside capital is realistically raisable at today's comparables. One number.
If the funding number exceeds the trough, the company has an operating problem and time to solve it.
If it does not, the company no longer has only a product decision in front of it. It has a decision about capital and strategic options, and the list of responses is longer than founders assume. Raise minority equity. Take debt. Slow the transition and stage it. Reprice packaging first and rebuild architecture later. Acquire capability rather than build it. Merge, so that one product investment is funded by three revenue bases. Partner. Or sell, which is one option among several and rarely the first.
Every one of those can be evaluated while the board still has time. Delay does not remove the arithmetic. It progressively removes the choices.
Cristina's board approved a roadmap in July and believed it had dealt with AI. What it had done was decline, without discussing it, to fund a transition it had never costed, in a business whose revenue was already contracting inside its existing customer base.
Nobody in the room voted for that. It was still the decision, and in time it becomes a decision about who owns the company, taken by default, at a moment chosen by somebody else.
Doing nothing is not the absence of a decision. It is the one option that gets exercised without a vote.
Five questions separate a board that is managing this from one that is hoping.
The fifth is the one most boards have never been asked.
If a material share of your revenue is attached to your customers' headcount, the useful question is not yet whether to sell. It is how much strategic freedom you still hold, and for how long.
The Inflexion Review is a defined engagement for founders and boards well before any transaction is contemplated. It quantifies the revenue exposed to declining labour units, models the transition economics and the gross-margin consequences, estimates the capital required and tests it against funding capacity, and assesses the available paths as future investors and acquirers are likely to assess them, setting out the conclusions in a board-ready options paper and financial model.
A transaction is only one possible conclusion. The better answer may be to invest, reprice, acquire, raise capital, or wait deliberately. The purpose is to make that choice while it is still yours to make.
To discuss whether a review would be useful, contact Hani Iskander directly at hani@cube.capital. Initial conversations are confidential and without obligation.
Cristina and the business described are illustrative. They are constructed from patterns common to Australian software companies and do not represent any single client. Companies named are described from published results and public statements.
Sell-side technology M&A advisory. Sydney, Australia.
© 2026 Cube Capital Pty Ltd. All rights reserved.
Cube Capital provides M&A advisory services to wholesale clients only. This website does not constitute financial product advice. Advisory services are provided in accordance with applicable Australian financial services law.