.png)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
How Should CEOs Approach Usage Based Pricing for Agentic SaaS Products
The central pricing question for agentic SaaS is not whether to charge for usage. It is what usage should mean to the buyer.
That distinction now separates durable pricing models from expensive experiments. A customer may accept paying for a support issue resolved, a qualified lead routed, or a workflow completed. The same customer will resist a bill based on tokens, model calls, or agent minutes that do not map to a business result. Meanwhile, a vendor that keeps a highly autonomous agent on an unlimited seat plan can discover too late that its best customers are also its least profitable.
CEOs therefore face a design problem with strategic consequences. The meter sets the growth path, the gross-margin ceiling, the sales motion, the forecast, and the customer’s willingness to expand. It also signals what the company believes it is selling: a tool that helps an employee, or a system that performs work.
Monetizely’s position is clear: usage based pricing should become the primary model only when an agent independently completes a buyer-recognized unit of work that can be measured without dispute. Until that point, companies should retain a seat or access fee as the primary commercial anchor and use consumption limits or credits to protect margins.
The market already shows that “usage” must match the agent’s real role
Leading vendors are not converging on one universal meter. They are matching their commercial model to the degree of human involvement, the scope of work, and the buyer’s ability to recognize value.
As of September 7, 2026, GitHub Copilot Business charges $19 per user per month and includes 1,900 AI credits per user. Cursor Teams starts at $40 per user per month, with included model usage and on-demand usage after the included amount is consumed. Both products remain anchored to the developer because the developer still directs, evaluates, and owns the work.
By contrast, Intercom bills Fin AI Agent primarily on outcomes. A resolved chat or email conversation costs $0.99, while a qualified sales prospect costs $9.99. Salesforce offers Agentforce at $0.10 per standard action through Flex Credits, while also listing $2 Help Agent resolutions for customer-facing support. Those models move closer to business work because the agent acts in systems and, in some cases, completes the customer’s request.
The evidence matters because it rejects a common executive shortcut: treating all AI products as if they should move immediately to outcome pricing.
| Vendor and product | Current published model, as of September 7, 2026 | What the customer is principally buying | What CEOs should learn |
|---|---|---|---|
| GitHub Copilot Business | $19 per user per month, with 1,900 AI credits per user; excess pooled usage is charged at $0.01 per credit. (docs.github.com) | A developer productivity tool | The seat remains credible when a human performs and approves the work. Credits limit cost exposure. |
| Cursor Teams | $40 per user per month; plans include model usage, and on-demand usage is billed after included usage is consumed. (cursor.com) | An engineering environment used by named professionals | A predictable seat fee can coexist with consumption protection when model costs vary by task and model. |
| Devin Teams | $40 per full seat per month, plus shared on-demand credits; flex seats are free and draw from the shared pool. (docs.devin.ai) | Access for regular users plus extra capacity for variable workloads | Agentic coding still needs a human-centered commercial anchor when review and task selection remain material. |
| Salesforce Agentforce | Flex Credits cost $500 per 100,000 credits; standard actions use 20 credits, or $0.10 per action. Salesforce also lists a $5 user license that requires Flex Credits. (salesforce.com) | Executed actions across business systems | Actions work when the buyer can observe and count a defined system operation, even if the full business outcome takes longer. |
| Intercom Fin AI Agent | $0.99 per resolution, procedure handoff, or disqualification; $9.99 per qualified prospect. (intercom.com) | A completed customer-support or sales result | Outcome pricing earns trust when the vendor defines success tightly and does not charge for failed or abandoned work. |
The pattern is not a market split between “old” seats and “new” usage. It is a progression from human access, to agent activity, to completed work.
The Agentic Monetization Spectrum, or AMS, gives CEOs a practical way to locate an agent before deciding what to charge for. It rates an agent on three dimensions: zero-human ability, meaning how little human work remains; operational domain, meaning whether the agent handles one task, a full workflow, or work across functions; and output/cost ratio, meaning whether the value created rises faster than inference and operating cost. An agent with substantial human involvement should remain close to seat pricing. An agent that completes work with little human intervention, covers a broader workflow, and creates value far above its compute cost can support an output or outcome meter.
For practical use, we score Small or Linear as 1, Medium or Inflecting as 2, and Large or Exponential as 3. A total score does not set a price. It identifies the strongest primary meter.
The AMS points to a hard commercial truth: an agent does not deserve outcome pricing because it uses a large language model. It deserves outcome pricing when the buyer experiences it as a substitute for completed human work.
A developer who asks an assistant to explain a codebase still buys a better tool. A support leader whose agent closes 100,000 customer cases buys operating capacity. The first sale should center on user access. The second should center on resolved demand.
The five pricing decisions must precede the price card
The strongest pricing teams resist the urge to start with a unit rate. They first decide what the company is trying to achieve and which customer it is trying to serve.
Monetizely’s 5-Step Pricing Framework, set out in Monetizing Agentic AI as well, sequences the work in five connected decisions: Goals and Segmentation, Packaging, Pricing Metric, Price Points, and Operationalization. The sequence matters because each decision constrains the next one. A company cannot choose a credible meter before it knows which buyer it serves, what that buyer needs, and how much of the agent’s capability belongs in the offer. Nor can it run a variable-price model without the data, billing logic, and customer controls to support it.
The table shows why pricing is not a late-stage finance task. The metric is the central decision, but it only works when the surrounding offer and operating system support it.
Tokens, model calls, agent minutes, and GPU time can be useful internal cost measures. They are rarely good primary SaaS pricing measures.
A buyer cannot easily forecast token use because prompts vary, model routing changes, workflows branch, and a task may require several retries. Nor can the buyer connect 1 million tokens to a customer case resolved, a compliance review completed, or a qualified lead delivered. Charging for technical activity transfers uncertainty from the vendor to the customer precisely when the vendor claims to be reducing customer effort.
Salesforce’s Flex Credits illustrate a better intermediate design. The company meters a discrete action, such as updating a record, summarizing a case, or executing a flow. That is not yet a full business outcome, but it is a visible operation that an administrator can count and audit. Salesforce’s published rate card assigns 20 Flex Credits to a standard action, while its pricing page lists $500 per 100,000 credits.
Intercom moves one step further. Its published definition of a billable Fin resolution requires an actual answer, excludes a conversation where Fin merely asks a clarifying question and receives no response, and removes a charge if a customer later returns to the same conversation for more help. That definition makes the result measurable enough to invoice.
| What the agent does in practice | Primary meter CEOs should choose | What must be true before charging | Meter CEOs should avoid as the primary basis |
|---|---|---|---|
| Assists an employee who remains responsible for the work | Seat | The named user is the enduring value anchor | Tokens or prompts |
| Executes a defined system operation inside a broader workflow | Action | The action is observable, standardized, and recorded | Agent session length |
| Closes a routine customer request without a human escalation | Resolution | Success and failure states are deterministic | Number of messages in a conversation |
| Routes a prospect that meets buyer-defined criteria | Qualification | Criteria and routing destination are agreed in advance | Number of leads scanned |
| Completes a repeatable end-to-end business process | Verified outcome | The customer can audit completion and exceptions | Model calls, compute minutes, or tool invocations |
The decision matrix makes the governing rule visible: move toward outcomes only as the buyer’s ability to recognize and verify completed work becomes stronger.
Usage based pricing does not require unpredictable revenue or surprise invoices. It requires a clear distinction between the meter and the buying commitment.
For an autonomous support agent, the primary meter may be a resolved case. The commercial agreement can still include an annual minimum purchase, monthly usage reporting, alert thresholds, and volume rates. The customer receives a budgeted commitment; the vendor retains expansion revenue when demand grows. The result is not two competing pricing models. It is one meter, resolved cases, purchased through an enterprise-friendly commitment.
Several current offers show the practical value of this architecture. Salesforce supports both pay-as-you-go and pre-commit purchasing for Flex Credits, and bills excess consumption at the contracted rate in arrears. Cursor includes a set amount of usage in its plans and allows on-demand billing after that amount is consumed. Devin’s Teams plan combines fixed full seats with a shared credit pool for variable work, while allowing administrators to control spend.
The CEO’s task is to decide which element carries the value story. For a mature autonomous workflow, the answer should be the outcome. A platform fee may cover access, security, deployment support, or baseline capacity, but it should not obscure the primary meter.
Billing credibility must be designed before the first enterprise rollout
An outcome meter creates more value alignment than a token meter, but it also creates a higher burden of proof. Product, finance, customer success, and legal must agree on the same definition of a billable event.
The operational standard should be higher than “we can log it.” The customer must be able to see the count, understand why it changed, and challenge an exception without opening a weeks-long investigation. Monetizely’s view is that an agentic pricing rollout should not reach general availability until those conditions are met.
The implication is straightforward: pricing operations are part of the product. If a company cannot explain a charge in a customer meeting, it has not finished designing the offer.
The largest mistake is to convert the whole installed base at once because agentic AI has become a board priority. A broad migration can damage adoption, distort usage data, and force sales teams to defend a meter that product has not yet earned.
A better path begins with one workflow where the agent already has three qualities: reliable completion, a clear system record, and a buyer-visible result. Customer support is often a good starting point because the work has a natural beginning and end. Lead qualification can work when the customer sets the criteria. Coding agents should generally wait longer, because human judgment still shapes task definition, review, testing, and deployment.
The pricing roadmap should also reflect capability gains. As an agent moves from assisting a user, to executing defined actions, to completing work with minimal review, the primary meter can move from seat, to action, to outcome. The company should make that change deliberately, with a new package and an explicit customer value story, rather than hiding a new meter inside a contract amendment.
Monetizely’s position is that CEOs should build agentic SaaS around one primary meter per workflow, not one fashionable meter for the whole portfolio. Outcomes should lead where agents truly finish work. Seats should lead where people remain the core source of judgment. Actions and credits belong in the middle, where the agent performs meaningful work but the final result still depends on several moving parts.
Choose the first workflow for outcome pricing based on proof, not visibility. Select the workflow with the cleanest completion record and the lowest dispute rate, even if it is less prominent than the company’s flagship feature.
Make the primary meter part of product positioning. Sales messaging, onboarding, dashboards, and executive business reviews should all describe the same unit of value, such as “resolved customer request” rather than “AI usage.”
Create a portfolio map using the AMS every quarter. Require product leaders to show which agents remain assistive, which have become workflow operators, and which have earned a transition to an outcome meter.
Set compensation around profitable customer expansion, not merely usage growth. Reward sales and customer-success teams for committed growth in billable outcomes that meet target gross-margin thresholds.
Retire pricing exceptions rather than accumulating them. Every exception should have an expiry date, a named owner, and a decision on whether it becomes a standard package rule or disappears at renewal.
Footnotes
- https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
- Salesforce, Agentforce pricing and Flex Credit billing materials, accessed September 7, 2026. (salesforce.com)
- Intercom, “Fin AI Agent outcomes,” published July 30, 2026. (intercom.com)
- GitHub, Copilot billing for organizations and enterprises, accessed September 7, 2026. (docs.github.com)
- Cursor and Cognition, current pricing and billing documentation, accessed September 7, 2026. (cursor.com)