Add preferred source
The enterprise software industry is shifting from flat monthly fees based on seat counts to billing based on AI usage and actual work outcomes. Salesforce has introduced the “Agentic Work Unit,” which counts AI work throughput, while Microsoft uses a credit-deduction method, and Fin and Zendesk operate outcome-based models that charge only when an inquiry is resolved. Gartner projects that by 2030 the enterprise application market, worth up to $234 billion, will be reshaped. The key question is how the industry will agree to measure and verify AI outcomes.
Key Elements
The pricing formula of the enterprise software (SaaS) industry is being shaken up. As the practice of charging flat monthly fees based on headcount and seat counts hits its limits with the arrival of AI agents, pricing based on usage and actual work outcomes is spreading rapidly.
Global customer relationship management (CRM) company Salesforce said in its second-quarter (May–July) earnings announcement in August that the volume of work handled by its AI agents reached 3.2 billion cases on an “Agentic Work Unit (AWU)” basis. AWU is Salesforce’s proprietary metric that counts the number of times AI actually performs work, such as responding to customer inquiries or updating information. In other words, the company that grew the enterprise SaaS market with a subscription model has officially put out numbers signaling that it will use “what AI did” rather than seat counts as its billing basis.
Market research firm Gartner projected that the spread of agentic AI will reshape the enterprise application market, worth up to $234 billion, by 2030. That amounts to roughly 20% of total enterprise SaaS spending.
From flat per-seat fees to usage and outcomes
Conventional business software pricing was tied to headcount and seat counts. But in an environment where a single employee runs multiple AIs simultaneously and consumes enormous computing power, it has become difficult to receive reasonable compensation under a per-seat flat-fee structure. Jason Lemkin, CEO of the SaaS founder community SaaStr, pointed out that as the number of AI agents that automatically perform work increases, the per-user billing formula itself is being shaken.
Companies are responding in two ways. First is layering a usage meter on top of per-seat subscription fees. When Microsoft launched its AI work tool “Copilot Cowork” in June, it began deducting credits by summing AI calls, data searches, tool use, and execution time. Salesforce charges $0.10 (~130 won) per action of its AI “Agentforce.”
A model that charges only when an outcome is confirmed has also emerged. Fin, an AI customer-service company Salesforce agreed to acquire in June, charges $0.99 (~1,300 won) per case when the AI produces a specific result, such as resolving an inquiry, and does not bill when resolution fails. The IT outlet The Information reported that OpenAI has also begun offering some large customers the option of paying only when AI successfully completes a specific task. Tomasz Tunguz, founder of Theory Ventures, stressed that pricing should be attached to the “work,” not to tokens or inference costs.
The sticking point is how to measure outcomes
As the billing unit shifts to outcomes, a new challenge has emerged: how to handle cases where a customer who was unsatisfied with the AI response simply closed the inquiry window. With no industry-wide standard yet, detailed conditions vary from company to company.
Zendesk uses another AI to verify the conversation the AI had with the customer and does not charge if it determines the issue was not satisfactorily resolved. Fin considers an outcome achieved if there is a positive response such as “this was helpful” or if the customer makes no further request. Salesforce’s “Agentforce Help Agent” recognizes a task as complete only when at least two messages have been exchanged with the customer, there is no negative feedback, and the session ends without being transferred to a human agent.
Outcome-based billing is still at an early stage. According to venture capital firm High Alpha, only 5% of companies that monetize AI apply outcome-based pricing. Jang Jin-seok, head of BCG X Korea—the AI technology design and build organization of BCG—said that for outcome-based models to spread, companies and providers must first agree on what counts as an outcome and how it will be measured and verified.
| Category | Billing basis | Representative examples |
|---|---|---|
| Traditional SaaS | Flat monthly fee based on seat count/headcount | Enterprise software generally |
| Usage-based | Credit deduction for AI calls, searches, execution time, etc. | Microsoft Copilot Cowork, Salesforce Agentforce ($0.10 per action) |
| Outcome-based | Charged only when a result occurs, such as inquiry resolution | Fin ($0.99 per case), Zendesk, OpenAI option for some large customers |
Note: The share of companies applying outcome-based billing is based on figures compiled by venture capital firm High Alpha.
Impact on the market landscape
As the billing axis shifts toward outcomes, the revenue recognition structure and earnings volatility of software companies will change as well. Usage-based models see revenue grow automatically as customer usage increases, whereas outcome-based models book revenue only after verification that the AI actually solved the problem. For providers, revenue predictability declines, but room opens up to charge a premium for products with proven performance.
For corporate procurement managers, it means a change in how costs are controlled. As the old cost-cutting strategy of reducing seat counts no longer works, deciding which tasks to hand to AI and how to measure the outcomes emerges as a new budget-management challenge. Industry observers say that whoever establishes the outcome-measurement standard first will determine pricing power in the enterprise AI market going forward.
