Usage-Based Pricing: Why SaaS Is Adding, Not Replacing

Usage-Based Pricing Isn't Replacing Seats in 2026 — It's Being Bolted Onto Them

Usage-based pricing isn't replacing seats in 2026 — it's being bolted onto them. Benchmarkit's survey of 342 SaaS companies found usage-priced businesses post 108% net revenue retention versus 98% for seat-priced ones, a real 10-point gap driving vendors like ServiceNow, Salesforce, and HubSpot to add consumption meters this year. But Ramp's analysis of spend across 200+ vendors shows consumption pricing is still just 4–6% of actual buyer spend, flat over the past twelve months. Vendors have moved fast. Customers have not moved at all.

The Shift Is Real, But It's Not the Shift Most Headlines Describe

The framing that "usage-based pricing is eating seat-based SaaS" implies a replacement already underway. The 2026 data tells a narrower story: almost every major vendor that changed its pricing this year added a second billing mechanism next to the seat rather than removing it. ServiceNow's CEO told investors on the company's Q2 FY26 earnings call that half of its net-new business is now sold outside the traditional per-seat model — while also confirming the company is keeping seats in place because customers want a predictable number to budget against. That's the pattern across nearly the entire market: vendors are adding a meter, not deleting a model.

The reason this is happening now, rather than at any point in the last two decades of SaaS, comes down to one structural break: AI agents decouple software usage from headcount. A team of 50 using a CRM can shrink to a team of 10 with AI agents handling the rest of the workload — while the software itself gets used more, not less. Charging per seat in that scenario means a vendor's revenue falls exactly when its product is delivering more value, which is backwards from how pricing is supposed to work.

What "Usage-Based" Actually Turned Into: The Credit, Not the Outcome

Much of the 2026 commentary on SaaS pricing focused on outcome-based billing — charging per resolved support ticket or qualified lead rather than per seat. Outcome pricing generated the most attention, but a different model actually spread across the industry: the credit. Between June and July 2026, GitHub, OpenAI, and Anthropic all moved flagship products onto credit-based metering within about seven weeks of each other, without any coordination between the companies — a signal the industry was converging on a standard rather than one company copying another.

The mechanical reason credits won over both pure usage pricing and pure outcome pricing is straightforward. Raw per-token pricing is nearly impossible for a buyer to estimate in advance, which creates budget anxiety. Outcome pricing requires negotiating a precise definition of what counts as a resolved outcome, which only works at enterprise contract sizes large enough to support that negotiation. Credits sit in between: the vendor sells a number that feels fixed on a pricing page while metering real consumption underneath it.

The rate cards that emerged in 2026 tell a consistent story. Salesforce priced Agentforce at $2.00, or 400 Flex Credits, per resolution regardless of how many actions the agent takes internally — credits sold at $500 per 100,000 and shared across actions, prompts, translations, and voice. HubSpot cut its Breeze Customer Agent price from $1.00 to $0.50 per resolved conversation in April 2026, alongside a $1 charge per recommended lead — a repricing that led to a guidance hit the same quarter it shipped. GitHub Copilot replaced its "premium request unit" system with AI Credits tied to real token consumption in June 2026, while leaving seat prices for Pro, Business, and Enterprise plans unchanged. monday.com ran a deliberate two-year migration: giving every paid tier an identical, revenue-losing allowance of free credits for roughly a year purely to collect consumption data, before shipping paid credit bundles alongside seats in May 2026 at a price offset by a matching discount, so no customer's bill jumped on the day the model changed. Docusign is the rare case that actually removed the seat: its Enterprise IAM product moved to pure consumption in June 2026, with unlimited users and no per-seat charge at all — aimed at workflows where the people generating the most usage on the platform were often people who never logged in.

The Number That Actually Justifies the Shift: 108% vs 98%

The single strongest data point behind the push toward usage pricing comes from Benchmarkit's 2026 annual metrics report, which surveyed 342 SaaS companies and found seat-based companies average 98% net revenue retention against 108% for usage-based companies. On $10 million of existing revenue, that's the difference between last year's customers paying $9.8 million this year versus $10.8 million — a full $1 million swing from the same customer base, driven entirely by how the vendor counts and bills.

That gap comes with two caveats that get lost when the number is repeated without them. First, it's a correlation across a mixed sample, not a controlled experiment — usage-priced companies skew toward infrastructure and AI-native categories that were already growing faster for reasons unrelated to their pricing model, so switching a pricing model doesn't automatically import that growth rate. Second, and less frequently cited, the same Benchmarkit report found gross revenue retention across B2B SaaS fell from a median of 88% to 84% over the same period, even as efficiency metrics hit a five-year high — meaning the customer base is leaking faster underneath while expansion revenue works harder to compensate. Usage-based pricing amplifies both directions: it grows automatically when a customer's consumption rises, and it shrinks automatically, with no floor, when consumption falls — which is precisely the risk seat pricing didn't carry.

Investors Have Already Made Their Preference Clear

The clearest signal of where the market believes revenue durability actually comes from shows up in investor preference data rather than vendor announcements: among investors surveyed as part of the same benchmark data, only 5% said they prefer seat-based pricing in a company they're evaluating, against 35% for hybrid, 26% for outcome-based, and 24% for usage-based models. That preference translates directly into valuation. Public SaaS companies with AI-native, usage- or outcome-priced revenue have been trading at 8x to 25x revenue in 2026, while seat-priced horizontal applications and front-office CRM software have compressed to 1.8x to 4.0x — meaning the same $20 million of annual recurring revenue can be worth anywhere from roughly $36 million to $500 million depending purely on which pricing category a buyer places the company in.

That gap showed up in an actual transaction: Airtable, a seat-priced collaboration platform, agreed to sell to Bending Spoons in early August 2026 at a valuation representing roughly a tenth of its peak — in the same week that Atlassian, also a seat-priced company, beat earnings guidance and its stock rose more than 20%. The lesson isn't that the market is punishing seats as a category; it's that the market is pricing in whether a company's revenue is expected to expand, and pricing model is the most visible public signal of that expectation.

The Market Event That Made This Urgent: February 2026's "SaaSpocalypse"

The abstract argument about seat compression became concrete in a single 48-hour window. On February 3, 2026, Anthropic's launch of Claude Cowork — an agentic tool capable of automating legal, financial, and multi-step enterprise workflows — triggered a selloff that erased approximately $285 billion in software-sector market capitalization, according to reporting from Bloomberg and other financial outlets. Traders at Jefferies reportedly coined the term "SaaSpocalypse" to describe the panic. The underlying investor logic was direct: if AI agents can do the work that justified a company buying ten software seats, the seat-based revenue model those companies were valued on looks structurally weaker than it did a week earlier.

The debate over how much that reaction reflected reality versus overcorrection continued for months. Box CEO Aaron Levie told CNBC in early August that the market had built a mistaken thesis around agents being uniformly bad for software categories over the prior six months, arguing that many investors had misread the situation. That same week, HubSpot's stock fell after an earnings report in the same quarter it had shipped outcome-based pricing on its Breeze agents, while Atlassian — still selling almost entirely by seat — beat guidance with 28% year-over-year revenue growth and its stock rose sharply. The market's actual read in 2026 turned out to be more selective than "seats are dead": it's punishing revenue it doesn't believe will expand, not the seat as a billing mechanism specifically.

The Buyer Data Nobody's Pricing Page Admits

The most underreported number in this entire shift comes from the buyer side rather than the vendor side. Ramp's analysis of spend across more than 200 vendors — including Salesforce, Atlassian, HubSpot, Adobe, and Oracle — found seat-based contracts still account for 65% to 75% of actual software spend, with consumption-based pricing stuck at just 4% to 6%, essentially unchanged over the prior twelve months. Adobe's billing remains 99% seat-based. HubSpot's own financial filings show usage-related revenue, including Breeze, at under 2% of total revenue.

The reason for that lag is structural rather than a sign vendors misjudged the shift: pricing pages can change in an afternoon, but enterprise contracts change at renewal, procurement changes on a budget cycle, and finance teams change only when someone forces the issue. That built-in delay of two to three years means a vendor announcing a new pricing model in 2026 is effectively pricing for buyers who won't be contractually ready to adopt it until 2028. Even at the largest seat-priced vendor in the market, seats are still growing: Microsoft 365 Copilot went from roughly 20 million paid seats in April 2026 to more than 30 million by the June quarter — though a leaked internal memo reportedly put actual weekly-active usage at only 20–30% of licensed seats, a reminder that seat counts and real engagement aren't the same signal.

Where the Credit Model Breaks: The Cursor Lesson

Usage-based and credit pricing carry a failure mode seat pricing never had: customers can't easily predict their bill in advance, and when the underlying cost assumptions change, the surprise lands on an invoice rather than a renewal conversation. In June 2026, Cursor changed its $20 Pro plan from a fixed 500 fast requests to "$20 of usage at API rates." Newer AI models consumed far more tokens per complex request than customers expected, and users who ran a handful of long, complex prompts burned through their allowance and received unexpectedly large bills. Cursor's CEO issued a public apology within weeks and refunded charges from a roughly three-week window. The underlying pricing logic was defensible — the company had been absorbing rising per-request costs that newer models created — but the credit abstraction meant customers couldn't see the change coming until it appeared on a statement, which is a materially different experience from a seat price increase announced ahead of a renewal.

What This Means for a Founder or Product Leader Right Now

Adding a usage meter doesn't require removing the seat. Nearly every successful 2026 migration — ServiceNow, Salesforce, HubSpot, monday.com — layered consumption pricing on top of an existing subscription rather than replacing it outright. The rare case that removed seats entirely, Docusign's Enterprise IAM, did so for a specific use case where per-user billing had stopped making sense, not as a general strategy.

Instrument consumption before you price it. monday.com spent roughly a year giving every tier an identical, non-monetized credit allowance specifically to observe real usage patterns before setting prices — the direct reason its eventual pricing launch didn't trigger a wave of surprise invoices or churn.

Decide what stays free, and publish it. Vendors that kept a clear line between metered and unmetered functionality — Gong's everyday AI features staying inside the base license while only background agent work draws credits — avoided the trust erosion that comes from customers feeling nickel-and-dimed on every interaction.

The buyer side is lagging the vendor side by design, not confusion. If your target customers are large enterprises, expect a two-to-three-year gap between announcing a usage-based option and it representing meaningful realized revenue, driven by renewal cycles and procurement processes rather than resistance to the model itself.