The evidence, and its limits.
Six figures sit behind our view. Each is quoted as published, with its date. Each is followed by what it does not show, because stretching a number past what it can support costs more than it’s worth. The first four are about the market. The last two are from the sectors we have written about.

The market is repricing, not shrinking
“SaaS will not be destroyed; it will emerge in a different form. This metamorphosis represents threats and opportunities for both incumbents and new challengers.”
“up to $234 billion of enterprise application spending exposed to agentic arbitrage between now and 2030”, about 20% of what companies spend on business software by 2030.
Gartner, press release, 1 July 2026. George Brocklehurst, Managing Vice President.
A fifth of business software spending sits in pricing that stops working when AI delivers the result directly.
That the money disappears. Gartner says the opposite, and against a category it separately expects to keep growing, this exposure could play out in full while the market still grows. It is a claim about which suppliers keep the money, which is why it has to be answered company by company.
Charging per seat is under pressure
“Seat-based pricing is under structural pressure.”
“NRR can mask seat contraction beneath expansion revenue.”
“AI lowers friction to build.”
PwC, “How AI is reshaping software valuations in M&A,” 24 February 2026.
How suppliers charge, whether retention figures flatter, and whether a customer could build it themselves are now mainstream diligence questions rather than contrarian ones. We credit PwC for the framing of workflow gravity and proprietary context.
A specific number. It is a way of thinking rather than a measurement. Turning that framing into a dollar figure for your own revenue is where we can help.
Where the substitution shows up first
“Employment of young workers (ages 22–25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers.”
“We find no evidence of widespread, economy-wide job displacement.”
Brynjolfsson, Chandar and Chen, “Canaries in the Coal Mine?” Stanford Digital Economy Lab, August 2026. Based on payroll records.
Substitution shows up first at entry level, in work that follows set rules, and it arrives as slower hiring rather than layoffs. If you bill against your customer's headcount in those roles, your clock started before their headcount fell.
Widespread job losses. The authors say plainly they find none, and that these are observations rather than proof of cause. Employment across the whole sample rose about 6%, and the most-exposed group across all ages grew about 4%. The statistical version of the headline roughly halves once you account for how much education a job requires, and stops being significant under full controls. We use it for where the change appears, not for how much.
Retention is slipping across the market
Median gross revenue retention fell from 88% to 84%; the top quartile fell from 95% to 91%.
Benchmarkit, 2026 B2B SaaS & AI-Native Metrics Report, covering 2022–2025.
Retention is softening widely enough that a board should test whether its own losses are execution or exposure.
The report's headline comparison between usage-based and seat-based retention. The same publisher's 2024 edition reported that relationship the other way round, the panel is rebuilt each year, and neither the method nor the group sizes are published. We treat it as something to test against a client's own numbers, not as a fact.
From the sector papers
Your revenue, described as someone else's market
the approximately $100 billion in annual revenue cycle labor services performed across the industry;
we believe we are well-positioned to automate a meaningful portion of this labor pool through new AI-powered capability launches like denials, prior authorization and recoupment.Matt Hawkins, Chief Executive, Waystar. Q1 2026 earnings, reported by Fierce Healthcare, April 30, 2026.
This is the other side of the trade describing your revenue as its addressable market, publicly, with a number attached. It is not a forecast. It is a business plan, and the company reports that roughly 40% of its revenue already comes from AI-embedded workflows.
How much of that $100 billion actually moves, or when. A vendors own addressable-market claim on an earnings call is the most self-serving genre in finance. And automation on one side of an adjudicated transaction has repeatedly manufactured work on the other: the American Hospital Association reports hospitals spent nearly $18 billion in 2025 on overturning claim denials alone. Labor statistics cut the same way. Medical transcription is projected to fall 4% this decade, while medical records and coding is projected to grow 8%.
Not every exposure is about AI
Commercial property and casualty premiums fell 1.2% in the first quarter of 2026, ending thirty-three consecutive quarterly increases. The second quarter fell 2.0%, with commercial property down 6.3%.Council of Insurance Agents & Brokers, Commercial Property & Casualty Market Index, Q1 and Q2 2026, reported by Insurance Journal and Risk & Insurance.
Commission is a percentage of premium. When the rate falls, the revenue falls with it, and nobody renegotiates anything. A brokerage can be excellent, retain every client, and still shrink. That exposure sits entirely outside the customer relationship, and no AI assessment will find it.
That this is the whole story, or that the series is sound. CIAB is a self-reported survey of its own member brokers. It publishes neither sample size nor methodology, and it defines account size by annual commission rather than premium. It is the best available series for this market and it is not a good one. Rate is also only one of four clocks: exposure units, automation and producer movement run on their own schedules, and in 2026 producer departures cost one public broker a projected $50 to $60 million.
