SAP is shifting toward AI consumption-based pricing, a move that signals how autonomous agents threaten the seat-based SaaS revenue model, ERP Today reports.

The logic is structural. Subscription software earns revenue per named user. AI agents do the work of users, so a customer that once bought 5,000 seats may need far fewer when agents handle transactions, reporting and data entry. Consumption pricing — charging for what is used, such as API calls, agent actions or compute — replaces the lost seat revenue with usage volume.

What does consumption pricing change for buyers?

For enterprise customers, the tradeoff cuts both ways. Costs scale down when usage falls, which helps teams that deploy AI selectively. Costs also become variable and harder to budget, a shift that finance departments accustomed to fixed annual license fees will have to model.

The model works best where usage is measurable and attributable to outcomes. It fails where vendors cannot define a clean unit of consumption, or where customers face unpredictable spikes that turn a cost-saving architecture into a cost overrun.

Why do agents threaten seat-based SaaS?

Agents break the assumption behind per-seat pricing: that every human worker needs a license. When one employee supervises dozens of agents, the ratio of licenses to work done collapses. Vendors across the enterprise software market, not only SAP, face the same arithmetic.

The failure mode for vendors is a transition gap: revenue from seats declines faster than consumption revenue grows. The failure mode for customers is vendor lock-in through proprietary metering, where usage cannot be independently verified.

What should product managers take from this?

Pricing is now a product decision, not a finance decision. Teams building AI features must instrument usage from the first release, define the billable unit early, and test whether customers can forecast their spend. Seat-based incumbents that delay the shift risk watching agent-driven workloads migrate to platforms priced for them.

The direction of travel is clear: as agents take on more execution work, expect consumption models to spread from hyperscalers into mainstream enterprise software.