AI pricing models are the frameworks enterprises use to buy artificial intelligence - today dominated by per-token and per-seat metering that prices compute and access but never the intelligence itself or the outcome it produces. Neither metric measures a fixed amount of useful thinking, which is exactly why procurement teams struggle to connect AI spend to a measurable return.
Every enterprise AI engagement we have observed this year has hit the same roadblock: procurement. Organizations are struggling with AI pricing models that feel fundamentally misaligned with the value being delivered. When procurement teams ask whether they should pay per seat or per token, they are often asking the wrong question. These metrics are relics of a previous era - tokens measure raw compute, and seats measure traditional SaaS access. The challenge for operations leaders is shifting from purchasing tools to managing a new economic input: synthetic intelligence.
Our research suggests organizations are caught in a cycle of underinvestment because they lack the framework to value this new input. While tokens are useful for metering information processing, they do not deliver a fixed amount of useful thinking - the burden of transforming raw processing power into a business outcome falls entirely on the buyer. For most mid-market companies, this creates a massive gap between the technical potential of a model and the actual return on investment achieved. It mirrors the broader problem of AI spend that chases tokens instead of outcomes.
Why AI pricing models built on tokens and seats fail
For centuries, business leaders have balanced two primary economic inputs - capital and labor. We are now witnessing the arrival of a third: intelligence. This is not just a faster way to do old tasks; it is a fundamental shift in how value is created. Consider Cedar, a healthcare payments company that runs an AI agent named Cora. Patients can call Cora at any time to have a medical bill explained, verify insurance, and settle charges on the spot.
In this scenario, the hospital does not care how many tokens were consumed during the 2:00 AM phone call. They care that the bill was paid and the patient was satisfied without waiting for a human operator. When we price this interaction per token or per seat, we treat the technology as a utility rather than a contributor to the balance sheet. This traditional product thinking is why so many CFOs find AI spending hard to justify: they can see the cost (the tokens), but they cannot easily allocate the intelligence to a specific return.
<!-- INFOGRAPHIC: side-by-side comparison of three AI pricing models - per-seat (meters access), per-token (meters compute), and outcome-based (meters resolved results) - showing which one a CFO can tie directly to ROI -->This lack of AI fluency - the ability of an organization to convert raw model capabilities into functioning business systems - is the root cause of the current ROI struggle. A model can look spectacular in a demo, yet the organization remains poorly equipped to use it. Without a clear bridge between the technology and the operational outcome, AI remains an experimental cost center rather than a strategic asset.
AI as a new economic input: synthetic labor
When we view AI as a new economic input, the conversation shifts from "how much does this tool cost?" to "what work is this intelligence performing?" This is the core of the solution-first model. Instead of buying a platform and hoping your team figures out how to use it, organizations are beginning to demand specific outcomes.
We are seeing outcome-based pricing emerge in specialized sectors. Customer service platforms, for instance, are starting to charge per resolved case rather than per agent seat - aligning the vendor's incentives with the client's goals. If the AI solves the problem, the vendor gets paid. This model is highly effective for closed-loop tasks like billing inquiries or technical support, the same pattern behind a governed AI customer support agent that is measured on resolutions, not logins. It is part of a larger shift toward an outcome economy for AI.
However, for more complex operational challenges - redesigning a manufacturing service program, optimizing a global supply chain - outcome pricing is harder to define. In these cases the value is found in how the intelligence is managed, and the focus shifts from unit cost to strategic alignment of the system. This is where the distinction between shadow AI and sovereign AI becomes critical. Shadow AI is a collection of fragmented experiments where employees use personal accounts to summarize emails; sovereign AI consists of governed, reliable systems an organization owns and controls to drive specific, measurable KPIs. Leaders who let the first take hold inherit a shadow AI governance crisis.
The rise of the manager of intelligence on the front lines
One of the most profound shifts our research has identified is the changing role of the individual contributor. Every employee is now, in effect, a manager of intelligence. Even without a management title, individuals decide which tasks to assign to an AI, how to verify the output, and when a human needs to intervene.
Consider an engineer at an equipment maintenance firm who notices a recurring failure pattern across multiple machines. Previously, diagnosing this might have taken weeks of manual record review. Now they can direct an AI agent to compare maintenance records and pull root-cause data. The engineer is not just a user; they are an allocator of intelligence, deciding where to dig deeper with the AI and where to stop and speak with a customer.
This bottom-up value creation is powerful, but it presents a risk to leadership. If intelligence management happens in a vacuum, the company loses the ability to scale those wins or secure the underlying data. Leadership must empower these frontline managers with professional-grade infrastructure rather than just access to a chat interface. The goal is to move from AI adoption - everyone has a login - to AI impact, where the company produces more with the same headcount, closing the adoption gap between tools and workflows.
From forward-deployed engineering to outcome-based pricing
To solve the AI fluency gap, major model providers have begun using "forward-deployed engineering" - specialized teams of engineers who embed within a client's organization to identify opportunities, design systems, and ensure they pay off in daily operations. Some have formed joint ventures to apply AI inside mid-sized manufacturing and healthcare systems.
While this model effectively addresses the fluency problem, it often resembles the massive, slow consulting projects of the past. For many mid-market and scaling companies, this approach is too slow and too expensive. It is why we advocate a Starter Project approach - a professional middle ground offering a fixed-scope, fixed-cost engagement that proves value within weeks. The economics of embedding engineers this way are explored further in our look at forward-deployed AI engineers.
By focusing on a single, high-impact operational outcome first, companies build internal AI fluency without the risk of an open-ended consulting contract. Once the first sovereign system is delivering ROI, the organization can expand into a long-term transformation partnership. This lets the CFO see a clear line between the investment and the result before committing to a company-wide overhaul.
Building AI fluency: a roadmap for operations leaders
How can an operations leader tell if an AI investment is actually working? We look to companies like Shopify and Block as benchmarks. Shopify has kept headcount roughly flat for nearly two years while revenue grew by 34%; its leadership monitors spending but prioritizes giving engineers the freedom to use the best models for the job. Block has similarly reported significant gross-profit growth by leaning into AI-driven productivity.
These companies are highly AI-fluent. They do not just buy tokens; they integrate intelligence into the heart of their business processes. For organizations not yet at this level, the roadmap involves three steps:
- Identify the unreached upside - instead of only cutting costs, look for revenue-generating projects you previously lacked the resources to tackle. Assign an AI-augmented team to those projects and measure the new revenue as a direct ROI of the intelligence investment.
- Standardize the stack - move away from fragmented tool use toward a governed system you own. Trinity, the operating system for AI agents, runs the reasoning, state, permissions, and audit trail in one place instead of scattering work across disconnected tools, and surfaces the numbers through an AI data analysis system so ROI is visible, not anecdotal.
- Empower the frontline - recognize your employees as managers of intelligence, and give them the training and sovereign tools to turn individual workflows into company-wide assets.
The bottom line
The era of buying AI like a utility is ending. As intelligence becomes a primary economic input, the winners will be those who stop obsessing over token costs and start focusing on operational outcomes. Whether you are a CEO looking to scale revenue without adding headcount or an individual contributor expanding the scope of your role, the path forward requires a shift in mindset: you are no longer just a user of technology, you are a manager of a new kind of labor. By closing the AI fluency gap and investing in governed, sovereign systems you own, organizations can finally move past the demo stage into a future where AI delivers consistent, measurable enterprise value.
