Per-seat licensing assumed one login equalled one unit of value. AI agents break that assumption because they act independently of any single human user, working around the clock without adding a seat. Gartner forecasts that 40 percent of enterprise SaaS offerings will include outcome-based pricing components by 2026, up from just 15 percent in 2022 a near-tripling in four years. Deloitte's TMT Predictions 2026 goes further, projecting that seat-based licensing's share of enterprise vendor revenue will fall from 21 percent to 15 percent as usage-, agent-, and outcome-based models absorb the difference. Both point to the same structural shift: pricing is decoupling from headcount and reattaching itself to consumption and results.
The clearest evidence sits in company financial filings rather than survey decks. Snowflake's fourth-quarter fiscal 2026 results, reported on February 25, 2026, showed product revenue of $1.23 billion, up 30 percent year-over-year, with remaining performance obligations reaching $9.77 billion, up 42 percent, and net revenue retention holding at 125 percent. Snowflake's own filings describe product revenue as inherently variable because it is recognised on platform consumption rather than contract duration customers can roll over unused capacity or exceed contracted volumes at will. Management has also flagged a structural tension particular to consumption models: performance improvements that lower customers' compute costs can simultaneously reduce the vendor's own revenue, a dynamic that never existed under flat subscription pricing.
Salesforce's Agentforce rollout illustrates how quickly vendors are iterating on consumption metrics. The product launched in late 2024 at a flat $2 per conversation. By May 2025, Salesforce had layered in Flex Credits, a consumption-based system priced at $500 per 100,000 credits, so that a standard AI action updating a record, running a workflow, resolving a case costs roughly $0.10, with per-user licences added later for teams with heavy, predictable usage. Zendesk has taken a similar path with outcome-based pricing, billing $1.50 to $2.00 per Automated Resolution a support ticket an AI agent closes without human intervention, confirmed after 72 hours of customer inactivity. These are not experimental pricing pages; they are the primary commercial architecture for two of enterprise software's largest customer-service platforms.
Consumption pricing solves a value-alignment problem for vendors but creates a forecasting problem for finance teams on both sides of the contract. Metronome's State of Usage-Based Pricing report found that 73 percent of SaaS companies running usage-based models are actively building processes to forecast variable revenue implying more than a quarter are still forecasting off incomplete data. On the buyer side, the Federal Reserve Banks of Richmond and Atlanta's CFO Survey, conducted with Duke University's Fuqua School of Business, found finance chiefs expecting broad increases in AI-related spending heading into 2026, even as they flagged pricing and cost pressure among their top concerns for the year. The mismatch is structural: usage-based revenue depends on post-signature customer behaviour, not a signed contract value, which breaks the pipeline-times-win-rate math finance teams have relied on for a decade.
Despite the forecasting friction, vendors are moving toward consumption pricing because the growth data supports it. Bessemer Venture Partners' AI Pricing and Monetization Playbook, published in February 2026 and drawn from a portfolio spanning more than 200 cloud companies, found that companies aligning revenue with customer outcomes are growing 1.5 to 3 times faster than peers still anchored to per-seat pricing. Speaking to CNBC in July 2026, Sierra co-founder Clay Bavor said the shift away from per-seat pricing is no longer theoretical, describing AI agents moving out of demos and into live customer service, sales, and support workflows where per-login billing simply stops making sense. Financial Times reporting, cited by PYMNTS in February 2026, described enterprise incumbents adapting consumption logic through credit pools and per-action billing so that revenue scales directly with the volume of work an AI system actually performs, rather than the number of employees who happen to have a login.
The direction of travel looks settled even if the exact mechanics are still being negotiated contract by contract. Billing platform Maxio reports that 83 percent of AI-native SaaS companies already offer usage-based pricing as a standard option, well ahead of legacy incumbents still mid-transition. For buyers, that means auditing every AI-linked SaaS contract for its underlying billing structure before the next renewal cycle, since a seat count negotiated today may cover a shrinking share of the actual AI-driven work by the time that contract comes up again. For vendors, the lesson from Snowflake's, Salesforce's, and Zendesk's disclosures is the same: usage-based billing rewards products that visibly save customers money or time, and punishes pricing models that were never really measuring value in the first place.