Salesforce tests outcome-based AI pricing alongside seats and Flex Credits

Salesforce is moving away from a pure per-user licensing model because AI agents and APIs are performing work that previously required named human seats. The company is testing a hybrid approach that includes traditional seats, consumption-based Flex Credits, and fees tied to specific outcomes, such as resolved customer service cases. This shift is driven by the fact that license revenue is currently a headwind for growth, while AI adoption complicates the logic of charging for human users.

The primary technical and operational challenge lies in defining what constitutes a successful outcome. Executive Vice President Bill Patterson acknowledged that agents often operate across multiple domains, making it difficult to isolate a single measurable result. Deputy CFO Mike Spencer noted that outcome-based pricing requires very objective measurements to drive the fees, a process that is inherently challenging to standardize across different customer environments and use cases.

For engineering and procurement teams, this transition introduces significant uncertainty into long-term budgeting. The company is offering various contract structures, including Agentic Enterprise License Agreements (AELAs) and the newer Salesforce Commit, which allows customers to commit to a total spend amount and allocate it across different consumption types. However, analysts have warned that all-you-can-eat agreements may not be available at renewal, forcing customers to negotiate limits on price increases if they move to defined usage allowances.

The current state of Salesforce's pricing is described internally as an 'anxiety-filled architecture' due to the coexistence of these different frameworks. While consumption revenue is expected to grow as more customers put AI systems into production, Spencer indicated it will take three to five years for this to become a material share of total revenue. In the interim, customers must navigate a complex landscape of contracts where the cost of AI adoption is not yet standardized or predictable.