Key Takeaways
|
While enterprise adoption of Generative AI has reached record levels, corporate boardrooms are undergoing an AI ROI reality check. The initial wave of unbridled enthusiasm has given way to rigorous scrutiny from CFOs and board members demanding tangible, bottom-line returns from the millions being invested in AI.
Enterprise AI implementations can also carry significant costs. Customizing and deploying autonomous systems requires compute resources, API usage, specialized engineering talent, integration work, and ongoing maintenance. At the same time, while AI can deliver measurable productivity gains at the individual employee level, translating those micro-efficiencies into measurable P&L impact can be challenging.
A primary driver of this challenge is legacy SaaS monetization. Traditional per-seat or compute-consumption pricing models offer predictable ways to purchase software, but they can create a disconnect between software cost and customer value. Billing based primarily on hardware compute, API calls, or tokens can leave customers paying more as AI performs more work, regardless of whether that work produces a meaningful business outcome.
This is driving interest in new approaches to AI agent pricing. Startups like Sierra, founded by Bret Taylor and Clay Bavor, are pioneering outcome-based pricing for enterprise AI agents. Sierra describes its model as one in which customers pay when the software achieves specific, valuable outcomes. The shift poses a broader question for enterprise software: Should customers pay for the resources an AI agent consumes, or for the work it successfully completes?
Resolution-Based Pricing Models
Charging for AI based on successful actions or resolved tasks represents a significant shift in software economics. Traditional SaaS sells access to software tools. Resolution-based pricing instead attempts to sell the completed work performed by autonomous digital workers. This shift can realign economic incentives:
Vendor Alignment
In traditional SaaS, vendors generate revenue regardless of whether a particular interaction produces a successful outcome. Under outcome-based pricing, vendors have a stronger financial incentive to improve model accuracy, autonomy, and speed because revenue is tied more directly to successful outcomes. Sierra explicitly positions its pricing around this principle.
Customer ROI and Risk Mitigation
For enterprise buyers, outcome-based pricing can reduce the risk of paying for unused capacity or unsuccessful interactions. Costs scale with workload and defined outcomes rather than simply the number of seats or underlying compute resources. However, while resolution-based pricing can better align incentives, it introduces several operational challenges:
Defining a Successful Resolution
What counts as a resolved case? If an AI agent closes a ticket but the customer contacts support again shortly afterward, was the issue genuinely resolved?
Unpredictable Budgeting
Costs can vary with ticket volumes, seasonal demand, and other changes in workload, making forecasting more difficult than with a fixed subscription.
Avoidance of Complex Tasks
Vendors could potentially have an incentive to focus on simpler, high-success tasks while avoiding complex cases that are more difficult to resolve.
Limited Economies of Scale
Unlike traditional software licensing, where adding users may have a relatively small marginal cost, outcome-based pricing generally increases as the volume of work increases.
These challenges mean outcome-based pricing is not automatically the right model for every AI application. It works best when the outcome is clearly defined, measurable, and attributable to the AI agent. Sierra itself notes that outcome-based pricing becomes more complex when outcomes are difficult to attribute, and that blended models can make sense for some use cases.
Hybrid Pricing Models
A hybrid pricing model combines two or more billing structures, most commonly a predictable fixed fee with a variable charge tied to usage, actions, or outcomes.
Instead of choosing between rigid per-seat licensing and completely variable pay-as-you-go billing, hybrid models give software vendors and customers a middle ground that balances financial predictability with usage or performance-based pricing.
Common Types of Hybrid Pricing Models
Platform Fee + Consumption
The customer pays a base monthly or annual subscription for platform access, core software features, security, and administration. On top of that, they pay for usage metrics such as API calls, AI tokens, or data processed.
Base Subscription + Included Allowance + Overages
Similar to a mobile phone plan, the subscription includes a baseline quota of usage, such as 2,000 tasks per month. Customers pay additional charges when they exceed the included allowance.
Subscription + Prepaid Credit Pools
Customers pay a recurring subscription that includes a fixed bundle of credits. Different activities consume different numbers of credits. Additional credits can be purchased as needed.
Platform Base + Outcome or Resolution-Based Fees
A baseline subscription covers platform access, administration, and agent setup, while the variable component is tied to successful outcomes, such as a resolved support interaction.
This hybrid approach can be particularly useful when some AI activities have easily measurable outcomes while others are better measured through consumption.
Salesforce Agentforce Pricing Models
Salesforce provides an interesting example of how enterprise AI pricing is evolving. Agentforce currently offers consumption-based pricing through Flex Credits or Conversations, alongside per-user licensing options.
The important distinction is that Agentforce pricing is not simply based on the underlying LLM’s token consumption. Depending on the capability being used, Salesforce meters activities through actions, prompts, conversations, or licensing structures.
1. Flex Credits: Pay Per Action
Flex Credits are designed around the actions an Agentforce agent performs. Salesforce describes an action as a specific function executed by the agent, such as updating a record, answering a product inquiry, or executing a custom prompt or flow.
Salesforce currently lists Flex Credits at $500 per 100,000 credits, with standard Agentforce actions generally consuming 20 credits each. Voice actions have different credit multipliers. This model shifts the unit of pricing from raw tokens toward the work performed by the agent.
2. Conversations: Pay Per Conversation
Salesforce also offers conversation-based pricing, where customers pay based on customer-agent conversations rather than individual actions.
This model is particularly relevant for customer-facing use cases where the interaction itself is a useful and predictable unit of consumption. Salesforce currently lists Conversations at $2 per conversation in its pricing materials.
Unlike outcome-based pricing, however, conversation pricing does not necessarily require the conversation to produce a successful business outcome. The customer pays for the interaction itself.
3. Per-User Licensing
Salesforce also offers per-user licensing options for employee-facing Agentforce use cases. These models introduce a familiar SaaS pricing mechanism into an otherwise consumption-oriented AI environment.
This can make sense where the value of an AI agent is closely tied to employee access and adoption rather than a discrete business outcome.
4. A More Nuanced Approach to LLM Consumption
The underlying AI model still matters to Agentforce economics, but not always in the same way as a traditional token-based API bill.
Salesforce’s current billing framework includes different usage types for actions and prompts. Prompt-based usage can be metered according to calls to the LLM gateway, token thresholds, and the type of model being used. Salesforce’s 2026 rate card also distinguishes between Salesforce-managed models and Bring Your Own LLM scenarios.
This creates a layered pricing model: Software access → Agent actions → AI prompts → Underlying models
The result is not a pure outcome-based model. Instead, Agentforce illustrates how enterprise AI vendors are experimenting with multiple economic units at the same time: users, conversations, actions, prompts, credits, and business outcomes.
The Larger Shift in AI Software Economics
The movement from seat licenses and raw token consumption toward actions, outcomes, and hybrid models reflects a broader change in what enterprise software does.
Traditional software primarily helps people perform work. AI agents increasingly perform portions of that work themselves.
That distinction matters because the economic unit of software can change with it. When software acts more like a digital worker, pricing based purely on access to the software becomes less intuitive. Customers may increasingly ask:
- What work did the agent perform?
- How much of that work was completed successfully?
- What business outcome did it produce?
- How predictable are the costs?
- Who carries the financial risk when the agent fails?
This does not mean seat-based or consumption-based pricing will disappear. Different AI workloads have different characteristics, and some are easier to measure through usage than outcomes.
The emerging model is therefore likely to be more nuanced than simply “pay for outcomes.” Vendors may combine subscriptions, usage, actions, credits, conversations, and outcomes depending on how autonomous the AI is and how clearly its impact can be measured.
Conclusion
The transition from seat licenses and raw token consumption toward value-based billing marks an important evolution in enterprise software economics. As Generative AI shifts software from passive tools toward autonomous digital workers, traditional monetization models face a new challenge: the cost of software is becoming increasingly disconnected from the value of the work it performs.
Companies such as Sierra are pushing outcome-based pricing, while enterprise platforms such as Salesforce are introducing action-based and hybrid consumption models. Sierra explicitly ties its pricing to outcomes, while Salesforce’s Agentforce combines Flex Credits, conversation-based pricing, and per-user options.
The future of AI software monetization may therefore be less about choosing one pricing model and more about matching the economic unit of pricing to the unit of value. For some applications, that unit will remain a user or a subscription. For others, it may be a conversation, action, workflow, or successful resolution. The larger shift is from paying simply for access to software toward paying for the work that software performs and the value that work creates.
Frequently Asked Questions (FAQs)
- What is outcome-based AI pricing?
It is a pricing model where customers pay when an AI agent delivers a predefined business outcome, rather than simply for software access or AI usage. - How is outcome-based pricing different from token-based pricing?
Token-based pricing charges for AI processing. Outcome-based pricing charges based on whether a defined business result is achieved. - What is resolution-based pricing?
It is a form of outcome-based pricing where the billable unit is a successfully resolved task or interaction. - What are Salesforce Agentforce Flex Credits?
Flex Credits are Salesforce’s consumption-based pricing unit for Agentforce. They meter activities such as agent actions and other usage types. - Is Salesforce Agentforce outcome-based?
Not entirely. Agentforce combines Flex Credits, conversation-based pricing, and per-user licensing, making it a hybrid approach rather than a pure outcome-based model. - Will AI agents replace traditional SaaS pricing?
Not necessarily. Subscription, seat-based, consumption, and outcome-based models are likely to coexist depending on the use case and how value is measured.
