.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 Can Partners Monetize AI Agents? 8 Reselling Models for Agentic AI Capabilities
Partners face a commercial choice that is more consequential than adding AI agents to a services catalog. They must decide what the customer is buying from them: access to a vendor product, a configured workflow, or completed work. The distinction determines whether the partner earns a short implementation fee, a thin resale margin, or a recurring revenue stream that grows as the agent takes on more work.
The market already shows several answers. Microsoft 365 Copilot is sold at $30 per user per month on an annual subscription. Cursor lists Teams at $40 per user per month, with included usage and usage-based charges beyond those limits. Salesforce offers Agentforce through action credits, conversations, user licenses, and flat-fee access. Intercom charges $0.99 for a successful Fin outcome. These are not cosmetic pricing choices. They reflect different levels of human involvement, scope, and proof of value. -
Monetizely’s position is clear: partners should build their agentic AI business around a recurring managed service with a fixed platform fee and a verified-outcome charge as the primary meter. Seats, credits, and flat fees still have a role, but mainly as entry points for lower-autonomy tools or as supporting parts of the offer. They should not define the partner’s core value proposition.
The strongest partner offers sell accountable work rather than discounted access
A partner selling a Microsoft 365 Copilot license can create revenue, but the client still sees Microsoft as the product owner and the partner as a procurement path. A partner that configures an agent to verify order status, process returns, update CRM records, and resolve routine support requests occupies a different position. It is responsible for the workflow, the quality controls, the integrations, and the business result.
That difference matters because vendor resale margins tend to be visible and contestable. The customer can compare a $30 Copilot seat or a $40 Cursor Teams seat with a vendor price page in minutes. By contrast, a managed service that resolves verified customer issues combines software, operating knowledge, workflow design, testing, analytics, and ongoing improvement. The buyer is not purchasing a markup on tokens. The buyer is purchasing less work for the human team.
The right sequence begins before the pricing conversation. Monetizely’s 5-Step Pricing Framework moves from goals and segmentation, to packaging, to the pricing metric, then price points, and finally operationalizing the model. The sequence matters because a partner that picks a meter before choosing its target buyer will often price the technology it acquired rather than the work the customer wants done. As discussed in Monetizing Agentic AI, a small business seeking faster first-response times does not buy, evaluate, or govern an agent in the same way as a regulated enterprise seeking to automate a full service workflow. -, -
The five steps create practical discipline:
- Goals and segmentation establish whether the partner needs rapid adoption, higher margin, deeper account control, or a new recurring-revenue line.
- Packaging determines what each buyer receives beyond the agent itself, such as integrations, monitoring, security controls, reporting, and service levels.
- Pricing metric identifies the unit that most closely tracks customer value.
- Price points turn that unit into a rate that supports both customer economics and partner margin.
- Operationalization connects product telemetry, billing, invoices, contracts, and account management so the model can run every month.
The public cases of Cursor, Devin, Harvey, Sierra, and 11x point to the same lesson: packages work when they reflect what distinct customer groups need to accomplish, not when they merely divide a feature list into three tiers. -
The Agentic Monetization Spectrum, or AMS, provides the practical test for choosing the partner’s primary meter. It assesses an agent on three dimensions: zero-human ability, operational domain, and output/cost ratio. A score of 1 means small, 2 means medium, and 3 means large.
Zero-human ability measures how much human work remains. At the low end, the employee still does most of the work and the AI assists. At the high end, the agent performs the work and a person mainly reviews exceptions. Operational domain measures whether the agent handles one task, an end-to-end workflow in one function, or work across several functions. Output/cost ratio asks whether the business value rises roughly in line with compute cost, rises faster than cost, or rises dramatically faster than cost. The higher the combined score, the less defensible a seat-based model becomes.
Exhibit 1: AMS scoring shows which products can support a partner outcome model
| Product or product archetype | Zero-human ability | Operational domain | Output/cost ratio | Total | What the score suggests |
|---|---|---|---|---|---|
| Microsoft 365 Copilot | 1 | 3 | 2 | 6 | Human productivity tool - lead with seats and adoption services |
| Cursor | 2 | 2 | 2 | 6 | Developer workflow tool - seat-led offer with usage guardrails |
| Devin | 2 | 2 | 2 | 6 | Delegated coding work - capacity or work-package pricing can emerge |
| Harvey | 2 | 2 | 3 | 7 | High-value legal work - price can move beyond seats when outputs are proven |
| Sierra | 3 | 2 | 2 | 7 | Autonomous service workflows - outcome pricing is justified |
| 11x Alice | 2 | 2 | 2 | 6 | Sales workflow support - do not charge as though it replaces a full sales team |
| Zapier Agents | 2 | 2 | 1 | 5 | Activity-based pricing fits early workflow automation |
| Salesforce Agentforce | 2 | 2 | 2 | 6 | Action or credit pricing works while outcomes remain uneven |
| Intercom Fin | 3 | 2 | 2 | 7 | Verified-resolution pricing fits the customer value delivered |
| Zendesk AI agents | 3 | 2 | 2 | 7 | Verified-resolution pricing fits the customer value delivered |
| Oya live agents | 2 | 2 | 1 | 5 | Flat per-agent pricing can support predictable deployment, but not a mature outcome model |
Source: Monetizely assessment using the AMS dimensions. Product descriptions and public offers checked September 3, 2026. -, -
The table explains why a reseller should not treat all AI agents as one category. A score of 5 or 6 signals that the employee still anchors the value. A score of 7 signals that the agent itself is increasingly the unit of productive capacity.
The current market contains per-seat, activity-based, action-credit, outcome-based, and flat-price models. A partner should study them less as templates to copy and more as evidence of what each vendor can observe and bill for.
Exhibit 2: Public agentic AI pricing models, checked September 3, 2026
| Vendor | Product | Public model | Meter | Published price or structure | Source |
|---|---|---|---|---|---|
| Microsoft | Microsoft 365 Copilot | Per-seat subscription | Named user | $30 per user per month, paid yearly | |
| Cursor | Cursor Teams | Per-seat with usage beyond included limits | Named user and model usage | $40 per user per month for Teams; on-demand usage is billed after included usage | |
| Zapier | Zapier Agents | Subscription with activity allowance | Agent activities | Pro: $33.33 per month when billed annually, including 1,500 activities | |
| Salesforce | Agentforce | Consumption, conversation, seat, and flat-fee options | Actions, conversations, users | Flex Credits: $500 per 100,000 credits; 20 credits per action, or $0.10 per action | |
| Intercom | Fin AI Agent | Outcome-based | Successful outcome | $0.99 per outcome, including resolutions and selected workflow outcomes | |
| Zendesk | AI agents | Tiered outcome-based | Automated or verified resolutions | Resolution allowance applied across outcome tiers; model introduced May 18, 2026 | |
| Oya | Oya AI agents | Flat per live agent or pay-per-run | Live deployed agent | $250 per live agent per month, or $2,500 annually, under fair-use terms |
Source: official vendor pricing and product documentation, checked September 3, 2026.
The pattern is straightforward. Vendors charge per seat when the human remains central, per activity or action when the product is still best understood as automation, and per verified resolution when the agent has taken responsibility for a customer-facing job.
Partners do not need to reject the first seven models. They need to assign each one a limited role. The mistake is allowing the easiest model to sell in the first quarter to become the business model for the next three years.
Exhibit 3: Eight models for monetizing resold agentic AI capabilities
| # | Partner model | Typical customer payment | AMS score of the agent being sold | Role in the portfolio |
|---|---|---|---|---|
| 1 | Referral-led introduction | One-time referral fee from vendor arrangement | 5 | Lead generation, not a durable revenue engine |
| 2 | Fixed-fee agent assessment and build | One-time project fee | 5 | Paid entry point that proves a workflow |
| 3 | Authorized per-seat resale | Monthly or annual license per user | 5-6 | Best for copilots and human-led tools |
| 4 | Managed employee copilot program | Per-user service fee plus vendor license | 6 | Adoption, enablement, and governance offer |
| 5 | Prepaid activity pack | Monthly activity or task allowance | 5 | Early automation and controlled pilots |
| 6 | Action-credit resale plus managed operations | Credits, overages, and support retainer | 6 | Useful while workflow output remains inconsistent |
| 7 | Flat fee per live agent | Monthly charge for each deployed agent | 5-6 | Predictable deployment model for narrow, stable work |
| 8 | Managed verified-outcome subscription | Fixed platform fee plus verified outcomes | 7+ | Core recurring model for autonomous workflow agents |
The first two models are useful because they lower the buyer’s commitment. A referral establishes trust, while a fixed-fee assessment can fund discovery, integration planning, and the first workflow. Neither creates durable account control unless it leads to an ongoing operating role.
Per-seat resale remains appropriate for employee copilots. Microsoft 365 Copilot and Cursor both fit this logic because the person is still the visible worker. The agent may accelerate a report, a spreadsheet, or a code change, but the employee owns the decision and the final result. A partner can add material value through training, prompt libraries, security settings, adoption reporting, and executive governance.
The commercial limitation is clear. Cursor states that it does not authorize resellers or third-party sellers, which means a partner cannot assume every AI product can be purchased wholesale and marked up. The right to resell, the right to invoice, and the right to support the customer must be confirmed before a partner builds a margin model around any vendor.
Activity and action models sit in the middle. Zapier measures agent activities. Salesforce sells Agentforce Flex Credits for actions, while also offering conversation, user-license, and flat-fee options. These meters help a vendor protect margins while agent reliability varies from workflow to workflow. For partners, they are useful inputs to internal cost management, but weak as the main customer story. Few buyers wake up wanting to purchase 50,000 actions. They want fewer support tickets, faster quotes, cleaner CRM records, or more qualified meetings. -
Flat pricing per live agent has a place when the work is narrow, stable, and easy to forecast. Oya, for example, publishes a $250-per-live-agent monthly plan with included usage under fair-use terms. That structure makes procurement simple for a customer that wants a fixed invoice. Yet a flat rate becomes harder to sustain when one agent processes 200 requests a month and another processes 20,000. The partner takes all the volume risk while giving the buyer no reason to connect payment with business value.
Model 8 is different. It separates the parts of the service that must exist every month from the work that creates variable customer value. The fixed platform fee pays for the configured agent environment, integrations, monitoring, reporting, knowledge maintenance, and operating support. The verified-outcome charge pays for completed work. The outcome charge is the primary meter because it is the part that scales with the customer’s realized benefit.
Falling inference prices will compress cost-led resale margins
A partner that bills customers primarily for tokens or underlying model calls is effectively promising to pass through a commodity input at a markup. That margin will come under pressure as model providers lower prices, customers bring more workloads in-house, and competing partners choose cheaper models.
Public API pricing illustrates the direction. In March 2023, a major model provider listed GPT-4 prompt tokens at $30 per million tokens. By April 2025, its GPT-4.1 API listed input tokens at $2 per million. The two offerings are not identical products, but the commercial lesson is still decisive: infrastructure prices can move far faster than the value of an automated business result. -
A verified resolution does not become less valuable to a support organization simply because the cost of the underlying inference falls. The customer still avoids a human touch, reduces queue time, and gains around-the-clock coverage. Competitive pressure will affect outcome rates over time, but the discussion stays centered on the customer’s result rather than on the partner’s changing model costs.
Partners should therefore use vendor credits, tokens, and API charges as internal controls. They should not make those units the core invoice line for an autonomous agent. Cost should set a floor. Value and accountability should determine the customer-facing meter.
A fixed platform fee and verified outcomes create a disciplined customer contract
The recommended structure has two parts, each with a distinct purpose:
The contract must define the outcome before the agent goes live. Intercom counts an outcome when Fin resolves an issue, completes a workflow, or meets other defined criteria, and it charges only once per conversation. Zendesk has moved toward verified resolutions and tiered resolution allowances, reflecting the same need for a defensible measurement rule. -
A partner should maintain an auditable record for every billable outcome:
Without that record, outcome pricing turns into a quarterly argument. With it, the invoice becomes a performance report.
The next exhibit shows how a managed outcome subscription can make customer spend understandable without turning the partner into a token broker. The primary meter is the verified resolution, while the platform fee pays for the continuous work that makes those resolutions possible.
Exhibit 4: Three-year customer cost under a managed resolution subscription
| Monthly verified resolutions | Annual platform fee | Annual outcome charges | One-time build fee | Year 1 total | Three-year total cost |
|---|---|---|---|---|---|
| 2,000 | $36,000 | $24,000 | $25,000 | $85,000 | $205,000 |
| 2,750 | $36,000 | $33,000 | $25,000 | $94,000 | $232,000 |
| 4,000 | $36,000 | $48,000 | $25,000 | $109,000 | $277,000 |
The customer can see what it will actually pay over three years, while the partner earns more only when the agent completes more verified work. That is a better economic relationship than a flat monthly fee that hides volume risk or a token charge that makes the buyer study technical logs.
The AMS provides a practical decision rule.
Exhibit 5: The partner meter should move with autonomy and proof of value
A partner should not leap to outcome pricing for a weak agent that cannot consistently finish the work. That creates disputes and destroys trust. Equally, a partner should not keep selling seats after the agent has become the real producer. That leaves revenue on the table and trains the buyer to treat the partner as a low-margin license channel.
The threshold is not whether the agent uses a large language model. The threshold is whether the agent can complete a recognizable unit of work with limited human involvement and a clear record of success.
Partners that own the operating model will capture the durable revenue
The best long-term position is not “AI reseller.” It is operator of a defined business workflow. A support specialist can run an AI resolution service. A revenue operations partner can run lead qualification and CRM-update workflows. A finance transformation partner can run invoice exception triage. A legal technology partner can run controlled contract-review workflows.
That operating role gives the partner four assets that a simple reseller lacks:
- A repeatable implementation method for a defined customer problem.
- A benchmark for expected agent performance across similar accounts.
- A data set for improving prompts, workflows, escalation rules, and knowledge sources.
- A commercial basis for charging on outcomes the customer can verify.
The first deployment may begin as a paid assessment or a seat-based pilot. The strategic destination should still be a managed outcome subscription. Partners that stop at resale will compete on access. Partners that take responsibility for completed work will compete on performance.
Partners should make four decisions now
Choose one workflow where success can be measured without debate. Start with a bounded process such as customer issue resolution, order-status handling, lead qualification, or document intake. Avoid broad “AI transformation” offers that cannot produce a billable result.
Build the offer around one target segment, not every possible buyer. A 50-person ecommerce company and a global insurer may both need service automation, but they require different integrations, controls, buying paths, and price levels.
Set the partner organization up to own a workflow P&L. Product, implementation, customer success, and finance teams should share responsibility for agent performance, renewal health, and gross margin.
Make verified output the central proof point in every executive review. Report the number of successful outcomes, escalation rate, exception types, customer impact, and the trend over time. Those measures support both renewal and expansion.
Footnotes
- Monetizing Agentic AI, Amazon listing, accessed September 3, 2026: https://www.amazon.com/Monetizing-Agentic-AI-Handbook-Transformation/dp/B0H7Z13VKJ/
- Monetizely, “Cursor: Segments Understood, Capabilities Mapped Well,” accessed September 3, 2026: https://www.getmonetizely.com/monetizing-agentic-ai-book-saas/cursor-segments-understood-capabilities-mapped-well
- Monetizely, “Devin: Right Segments, Wrong-Sized Packages,” accessed September 3, 2026: https://www.getmonetizely.com/monetizing-agentic-ai-book-saas/devin-right-segments-wrong-sized-packages
- Monetizely, “Harvey AI: Built for the Top, Invisible to the Rest,” accessed September 3, 2026: https://www.getmonetizely.com/monetizing-agentic-ai-book-saas/harvey-ai-built-for-the-top-invisible-to-the-rest
- Monetizely, “Sierra AI: Three Segments, One Served,” accessed September 3, 2026: https://www.getmonetizely.com/monetizing-agentic-ai-book-saas/sierra-ai-three-segments-one-served
- Monetizely, “11x Alice: One Package That Fits No One,” accessed September 3, 2026: https://www.getmonetizely.com/monetizing-agentic-ai-book-saas/11x-alice-one-package-that-fits-no-one
- Cursor, “Pricing,” accessed September 3, 2026: https://cursor.com/pricing
- Microsoft, “Microsoft 365 Copilot Plans and Pricing,” accessed September 3, 2026: https://www.microsoft.com/en-us/microsoft-365-copilot/pricing/enterprise
- Zapier, “Agents Plans and Pricing,” accessed September 3, 2026: https://zapier.com/pricing?filter_by=type%3A
- Salesforce, “Agentforce Pricing,” accessed September 3, 2026: https://www.salesforce.com/agentforce/pricing/?bc=OTH
- Intercom, “Pricing,” accessed September 3, 2026: https://www.intercom.com/pricing
- Zendesk, “About Automated Resolution Tiers,” updated August 24, 2026: https://support.zendesk.com/hc/en-us/articles/9570369117338-About-automated-resolution-tiers
- Oya, “Pricing,” accessed September 3, 2026: https://getoya.ai/pricing
- OpenAI, “GPT-4,” published March 14, 2023: https://openai.com/index/gpt-4-research/
- OpenAI, “GPT-4.1 Model,” accessed September 3, 2026: https://developers.openai.com/api/docs/models/gpt-4.1