Per-Seat Pricing Fell to 15% of the Market. Here Is the Autopsy

Pure per-seat pricing did not fall from 21% to 15% of the SaaS market in twelve months. The pair of numbers is real. It comes from a Gartner prediction about seat-based vendor revenue share by 2030, reported in June 2026. Somewhere between the analyst note and the tenth restatement, a four year horizon became a year that had already happened.

Where the 15% number actually comes from

Per-seat pricing has not been measured at 15% of the SaaS market. Gartner predicts that seat-based vendor revenue share declines from 21% to 15% as at least 40% of enterprise SaaS spending moves to usage, agent or outcome pricing by 2030. That is a four year horizon, reported in June 2026, and it describes revenue rather than companies.

The original sentence is a prediction

The clearest published version sits in CIO's June 2026 piece on enterprise pricing, written by Grant Gross. One sentence carries both claims. At least 40% of enterprise SaaS spending shifts by 2030, with seat-based vendor revenue share declining from 21% to 15%.

Read it as written and there is no twelve month collapse in it. There is a single prediction with a single horizon, and the 15% is the far end of it.

There is a second problem worth naming. The hyperlink CIO attaches to that prediction points at a software agency's marketing blog rather than at a Gartner document. Even the trade press citation does not resolve to the source it names.

How a 2030 horizon became a twelve month collapse

Trace the restatements and the sentence mutates in a consistent direction. Each retelling drops the horizon and tightens the timeframe.

The SaaS CFO published the pair in May 2026 as pricing that "has already fallen from 21% to 15% of SaaS companies in just twelve months". It attributed the figure to Kyle Poyar's monetisation research at Growth Unhinged. I went and read that research. It does not contain that pair.

What Poyar's 2026 report does contain is a different 21%. It reports that the share of companies offering multiple pricing models rose to 29%, "up from 21% last year". A number about model breadth has been read as a number about seat pricing.

A third widely shared piece gives the same 15% and 21% to an unrelated survey. It pairs them with hybrid pricing at 41%, up from 27%. Those hybrid figures do not match the Growth Unhinged survey's 37% and 25% either. Three articles, three attributions, one pair of numbers.

Six restatements are not six sources

This is the failure mode that a citation count cannot catch. A statistic appearing in eight places reads like corroboration. Check the footnotes and all eight trace back to the same sentence.

My position on this is blunt. A number that has been repeated more often than it has been sourced should be treated as unverified until you find the document, whatever your own priors say about the trend it supports.

I have to apply that to our own work. We published the 21% to 15% pair in the analysis of the seat compression spiral, described as a fall over twelve months and sourced to "published 2026 monetisation research". That framing was wrong, and this post is the correction.

The honest gap in this table: I could not read Gartner's underlying note, which sits behind a subscription. Row one is reported through CIO, so the wording is theirs rather than the analyst's.

What the measured studies actually found

Correcting a statistic is worth nothing if you do not put a better one in its place. Two 2026 studies were done by people who counted something and said what they counted.

Hybrid is the number that moved

Kyle Poyar's 2026 State of B2B SaaS and AI Monetization report surveyed 230 software companies during April and May 2026. Hybrid pricing came in as the most popular single model at 37%, against 25% reported for twelve months earlier.

That is a 12 point move in a year, and it is larger than the move the miscited statistic describes. The interesting shift was hiding behind the wrong number.

Two details in the same survey matter more for an operator than the headline. Three in four software companies changed pricing or packaging within the last year. And 29% already sell AI credits, with a further 33% planning to introduce them inside 6 to 12 months.

The size split is the part nobody quotes. Companies under $5 million in revenue gravitated to flat fees at 37%, while companies above $150 million held per-seat at 29%. Per-seat is not disappearing at the top of the market. It is being defended there.

Per-seat is still the primary model for large vendors

AlixPartners reviewed 65 major SaaS companies and AI-native competitors in a piece published on 29 July 2026. Four had fully adopted outcome-based pricing. Of the remaining 61, more than half still rely primarily on per-seat models.

The firm framed the finding around risk rather than around labels, which is the more useful framing. In 53 of the 65 companies, the customer carries the loss if the software does not deliver. Eight share that risk. Four take it on.

Seven of the nine vendors above $5 billion in annual revenue keep the customer holding the risk. That is the opposite of what a market abandoning per-seat pricing would look like.

What actually broke the link between seats and value

The provenance problem does not make the underlying shift imaginary. It is well documented, and the best version of it also comes from Gartner.

In a press release dated 1 July 2026, the firm put up to $234 billion of enterprise application software spend at risk from agentic arbitrage between now and 2030. That is roughly 20% of enterprise application SaaS spending by that year.

Agentic arbitrage is a precise idea. When an agent completes a task across several systems, the human stops opening those systems. The work still happens. The session does not. Managing vice president George Brocklehurst put the consequence plainly: this breaks the link between user growth and revenue growth for many enterprise software vendors.

That is the mechanism the miscited statistic was reaching for, and it deserves the better citation. The same dynamic is why the interface itself is under pressure, which is the subject of the piece on post-interface software. It is also why point products are being pulled into agent ecosystems.

The incentive inversion is not in dispute

Where a product genuinely reduces the number of people a customer needs, per-seat pricing makes delivering value expensive for the vendor. The sales team ends up quietly hoping the product underperforms.

That structural problem is real and it does not depend on any market share figure being accurate. It is examined at length in the seat compression analysis, and nothing in this post weakens it.

What the evidence does weaken is the timeline. This is a pressure building across a decade of contracts, not a market that repriced itself last year while nobody was counting.

The market's answer was a meter bolted onto a seat

Every measured source converges on the same destination, and it is not the one the coverage promises. Vendors kept the seat and added a meter.

AlixPartners found 72% of its 65 companies had layered consumption in through AI credits or usage metering. Poyar found hybrid at 37% and rising fastest. Both describe an addition, not a replacement.

The commercial logic is simple. A base fee gives procurement a number it can approve, and the meter gives the vendor exposure to growth that no longer arrives through headcount. Neither side gets its ideal model and both sides can sign.

Why pure consumption lost

Buyers say they want to move. A Futurum Group survey covering the first half of 2026 found 43% of buyers preferring consumption-based models and 27% favouring outcome-based structures. Fewer than one in five still preferred classic per-user pricing.

Then look at what they choose when both options sit on the same order form. Poyar reports that roughly 80% of Decagon's customers pick its predictable per-conversation model over its outcome-based one.

Stated preference says outcomes. Revealed preference says predictability. When those two disagree, price against the second one.

The hybrids that work in market keep the variable component small and legible. Intercom prices its Fin agent at $0.99 per resolution on top of a $49 monthly base, which is an outcome metric wrapped in a subscription rather than a replacement for one.

The counter-evidence is stronger than the thesis

If per-seat pricing were collapsing, the largest agent business in enterprise software would be the last place you would find it. It is exactly where you find it.

Salesforce moved back toward the seat and grew 205%

Salesforce launched Agentforce on per-conversation pricing and retreated from it. Marc Benioff told The Register in December 2025 that the company started out "talking about [charging] so much per conversation" before customers pushed for more flexibility.

The replacement is the Agentic Enterprise License Agreement, introduced in October 2025 and described by Benioff as the preferred licensing option. It bundles a menu of options and reusable credits into a single package, and it is priced on the number of seats.

The result is not a rounding error. Salesforce reported $1.2 billion of Agentforce annual recurring revenue on 27 May 2026, up 205% year over year, with Agentforce and Data 360 together at $3.4 billion. The fastest scaling agent product in enterprise software is sold on a seat-anchored licence.

Read that alongside the rest of the platform's position in the assessment of Salesforce in the agent era. The pricing retreat and the growth are the same story. Enterprises buy what their finance function can forecast.

How to read a pricing statistic before you repeat it

The reason four studies can look contradictory is that they count four different things. Once the denominators are visible, most of the disagreement disappears.

Four questions get you most of the way. Who published it, and did they count or predict? What is the denominator, revenue or companies? What date does the number describe, and is the horizon in the future? And does the citation resolve to a document you can open?

The last one catches more errors than the other three combined. Most bad statistics in this category are not fabricated. They are accurately transcribed from something that said a different thing. The same discipline applies to any vendor benchmark you are handed, which is the argument in the piece on rationalising a bloated application estate.

Where this argument is weakest

Four honest problems with what I have written.

The first is the biggest. I could not read Gartner's underlying note, which is behind a subscription. Everything I say about the 21% and 15% figures rests on how CIO reported them. If Gartner also published a measured 2025 baseline, my correction is partly wrong and I would want to know.

The second is that being right about provenance is not the same as being right about direction. Every source in this post points the same way, which is away from pure per-seat pricing. I am arguing about the speed and the evidence quality, not about the destination.

The third is sample quality on both sides of my replacement evidence. The Growth Unhinged survey is 230 self-selecting respondents over two months, and companies with strong opinions about pricing are the ones who answer pricing surveys. AlixPartners covers 65 large vendors, a deliberate skew toward the incumbents slowest to change. Neither is a census, and the two disagree in the direction their samples predict.

The fourth is ours. This publication has no first-party pricing data, so I am auditing other people's numbers rather than producing better ones. That is a weaker contribution than a measured study. The cost of AI delivery underneath all of this is covered separately in the analysis of inference costs and margins.

Frequently asked questions

Where to start this week

Open the last pricing deck your team produced and find every external statistic in it. Click each citation. Whatever does not resolve to a document with a date and a denominator comes out of the deck, however much you like what it says.

Then run one calculation on your own account base. Take your 20 largest customers, and for each one compare seat count twelve months ago with seat count today, alongside what they pay. If seats are flat or falling while spend holds, you are not in the trap and you have time. If both are falling together, you have a value metric problem and a live number to size it with.

Related reading