The evidence, and its limits.
Four figures sit behind our view. Each is quoted as published, with its date. Each is followed by what it does not show, because we want each number used only for what it can honestly support.
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.