Cost obsession is over: Tokenomics and outcomes redefine how enterprises buy AI

This HFS Point of View is for CFOs, CIOs, and enterprise transformation leaders rethinking how they buy, price, and scale agentic AI as it moves from cost cutting to outcome-based value.

The conversation dominating the world of AI-driven enterprise transformations is how to move away from this monolithic obsession with cost toward the dream of outcome-driven partnerships, where investments are driven by real business value and not simply removing more headcount. Forget “mess for less” and start thinking “a lot less mess for a lot more value.”

The cold, hard reality is that every AI transformation story starts the same way. A CFO demands ROI justification. A CIO frames the business case around headcount reduction and process automation. A project team builds a proof-of-concept to demonstrate the productivity math. For a while, that method works. More than half (54%) of AI use cases at the proof-of-concept (POC) or pilot stages are still rooted in productivity and cost, according to HFS Research data covering 979 GenAI and agentic AI use cases collected over the last 12 months.

Production gets real as intelligence at scale takes over

When those POCs move into production and are tested under real enterprise conditions, a remarkable transformation occurs. Enterprises that survive the pilot phase and commit to scaling AI in production find that 81% of their outcomes are operating in the top three tiers of value creation: performance, personalization, and prediction. The cost-and-scale imperative that justified the investment gives way to something far more powerful: an intent-driven, outcome-first mindset that opens up entirely new business possibilities.

This shift is the story the 4Ps of agentic AI tell, and every enterprise leader needs to understand how the 4Ps impact how they buy, price, deploy, and think about intelligence at scale.

The agentic 4Ps are a framework for what agentic AI actually delivers

The agentic 4Ps are not aspirational categories. They are empirical outcomes observed across hundreds of live enterprise AI deployments. Here is what each one means:

• Productivity is where most enterprises start their agentic journey. They focus on speed, efficiency, and scale, automating or augmenting repetitive tasks so that fewer people are needed to do the same work. At the POC stage, productivity dominates at 54% of use cases, and it is the language of cost-center thinking and a legitimate starting point, but not a destination.
• Prediction uses data and AI to anticipate, forecast, and optimize decisions before they are made. Representing 17% of use cases during POCs, it begins to stretch thinking beyond automation and into intelligence. By the time organizations move to production, this category grows to 25%, as leaders recognize that anticipating demand, risk, churn, or opportunity is worth far more than trimming operational costs.
• Personalization tailors experiences and interactions to individual needs and context. It moves AI from a back-office efficiency tool to a front-office value creator. Personalization climbs from 21% in pilots to 29% in production, as enterprises realize that individualized customer and employee experiences can drive revenue, loyalty, and differentiation.
• Performance is the apex: strategic outcomes, growth, new business models, and competitive leadership. Only 8% of POC use cases reach this tier. In production, the figure leaps to 27%. That gap is not a coincidence. Performance-level outcomes require commitment, not just experimentation. They require leaders willing to move beyond the cost conversation entirely.

Exhibit 1: Productivity dominates AI pilots; performance, prediction, and personalization dominate production

Sample: 979 GenAI and Agentic AI use cases collected by HFS over the last 12 months
Source: HFS Research, 2026

Agentic AI triggers the Jevons paradox at enterprise scale, and most leaders are not prepared for it

William Stanley Jevons observed in 1865 that improvements in the efficiency of coal use did not reduce coal consumption; it dramatically increased it. The paradox is counterintuitive but empirically consistent. When a resource becomes cheaper and more efficient to use, demand for it grows faster than the efficiency gains reduce it.

When AI can execute multi-step workflows autonomously and orchestrate processes across systems without human intervention, and when the cost of deploying intelligence drops toward near-zero marginal cost, the rational enterprise response is not to use less intelligence, but to apply intelligence to every decision, customer interaction, process, and growth opportunity that was previously uneconomical to address.

The CFO who authorized the initial AI investment to cut 30%–40% of operational costs may find that the same infrastructure, deployed with an outcome mindset, is now powering a new revenue line, a real-time personalization engine, and a predictive risk model that prevents losses that were previously invisible. The intelligence does not shrink to meet a cost target; it expands to fill the opportunity space.

This dynamic explains why the shift from productivity, which drops from 54% in POC to 19% in production while performance, personalization, and prediction increase, is not just a maturing of deployment. This shift is the Jevons paradox at work. Enterprises that scale agentic AI discover that the appetite for intelligence grows faster than the cost of delivering it. The constraint was the cost of deploying it at scale, not the technology. Remove that constraint, and the use cases multiply.

The thirst for value drives self-reinforcing outcomes

The Jevons paradox explains the consumption dynamic, but a parallel organizational psychology, equally important and less discussed, is at work. Once enterprises taste genuine AI-driven outcomes, their appetite for more does not plateau, but accelerates.

Call it the thirst for value. The thirst begins in the moment a leadership team stops asking “Did we get the ROI we promised the board?” and starts asking “What else can we do with this?” It is the behavioral shift that separates enterprises that treat AI as a cost project from those that treat it as a growth platform.

Exhibit 2: The thirst for value pulls enterprises up the 4Ps stack toward performance

Source: HFS Research, 2026

The pattern is consistent across HFS Research’s production-stage cohort. Enterprises that achieve a meaningful outcome in one domain, a measurable reduction in customer churn, a demonstrable improvement in forecast accuracy, and a new revenue stream enabled by AI-driven personalization do not pull back and declare victory. They redeploy, expand, and push the boundary of what they were willing to attempt before because the evidence of what is possible has changed their sense of what is permissible.

This progression is fundamentally different from the organizational dynamic that governs cost-reduction programs. Cost programs have a finish line. You hit the target, close the project, and move on. Value creation programs do not have a finish line, because the definition of what counts as value keeps expanding as capability matures. The thirst for value is what drives enterprises up the 4Ps stack, from productivity through prediction and personalization toward performance, not as a planned roadmap but as an emergent consequence of what they discover along the way.

The leadership implication is significant. Organizations that design their AI programs around a fixed cost target will hit it and stop. Organizations that design around an open-ended value ambition will keep going. Once activated, the thirst for value is a strategic asset.

The death of per-seat pricing and the rise of tokenomics

Traditional software and services pricing was built on headcount logic. You paid per user, FTE, license, or seat. The implicit assumption was that value scaled with the number of humans doing the work. AI shatters that assumption entirely.

When an agentic system can execute thousands of tasks simultaneously, the concept of a seat becomes meaningless. You are not replacing a person; you are replacing a workflow, and workflows do not have seats.

“Tokenomics” is the pricing architecture that reflects this reality. In the context of enterprise AI services, tokenomics means pricing consumption based on the units of AI activity: tokens processed, decisions made, tasks executed, or outputs generated. Instead of paying a fixed monthly cost per user, an enterprise pays for the intelligence it consumes, much like paying for compute or electricity.

The unit economics are transparent, usage-based, and directly tied to activity volume rather than headcount. Tokenomics is still a consumption model; however, it creates the perfect bridge to ambitious outcome-based commercial models, where pricing is tied directly to the business results AI generates.

From tokenomics to outcome-based pricing: The more value you generate, the more intelligence you consume, and the price scales accordingly

Here is what that looks like in practice across the Services-as-Software™ landscape:

• Revenue share on customer acquisition. An AI-powered lead qualification and outreach system is priced as a percentage of the closed pipeline it generates rather than a flat fee for running the software. The service provider wins when the enterprise wins.
• Cost avoidance guarantees. An agentic compliance monitoring system prices itself against the regulatory fines and penalties it prevents. If the system fails to prevent a breach, the customer receives a credit. The commercial model is directly aligned with the risk outcome it is managing.
• SLA-linked process execution. A business process automation provider running accounts payable or invoice processing charges per transaction only when the transaction completes within the agreed time and accuracy thresholds. Errors or delays reduce the billable event. Quality becomes the price mechanism.
• Throughput-based healthcare delivery. A clinical documentation AI prices per completed, compliant clinical note, not per license. The hospital pays for output, not access. Productivity gains for clinicians are measurable, and the pricing reflects them directly.
• Dynamic pricing on customer retention. An AI churn-prediction and intervention system charges a fee per customer successfully retained, measured against a defined baseline churn rate. The commercial incentive is to save customers, not to run software.

These models are not hypothetical. They are beginning to appear in enterprise contracts today, and they will become the standard commercial architecture of Services-as-Software as the market matures. The era of buying services and software by the seat is ending. The era of buying outcomes by the result has begun.

From operational obsession to intent-driven leadership

The shift from cost-based to outcome-based AI is primarily a leadership journey, not a technology journey.

Leaders who are wired for operational efficiency ask, “How do I do the same thing cheaper and faster?” Leaders who operate from an intent-driven mindset ask, “What becomes possible that was impossible before, and what new value can I create if I remove the constraints of cost and scale?”

These are fundamentally different questions. The first question keeps you inside the existing business model. The second question opens the door to entirely new ones.

Consider what happens when a retail enterprise stops asking, “How do I reduce contact center headcount?” and starts asking, “What if every customer interaction was personalized in real time to their context, history, and emotional state?” The first question leads to automation. The second leads to a competitive differentiator. The technology is the same, but the leadership mindset determines which outcome you pursue.

Intent-driven leaders do three things differently

Exhibit 3: Intent-driven leaders unlock growth with outcome focus, value measurement, and motivation

Source: HFS Research, 2026

1. First, they define outcomes before technology. They start with “We want to reduce customer churn by 15% in 18 months,” and work backward to the AI capabilities needed to deliver it, rather than starting with an AI capability and searching for a problem to solve.
2. Second, they measure value creation, not just cost reduction. They track revenue generated, risk avoided, decisions accelerated, and customers retained, alongside the operational savings that originally justified the investment. Basing decisions on value creation changes what gets funded and what gets killed.
3. Third, they motivate their teams around possibility, not compliance. When leaders articulate ambitious outcome targets, they activate a different kind of organizational energy. People want to solve meaningful problems. A target of reducing FTE cost by 12% is a managed decline. A target of helping 10 million customers make better financial decisions this year is a mission.

The data supports this distinction. The enterprises in HFS Research’s production cohort that have reached the Performance tier of the 4Ps are not the ones that had the biggest AI budgets. They are the ones with leaders who asked harder questions earlier.

Real enterprises are making the shift from cost models to outcome models

The pattern is consistent across industries. Here are the archetypes that characterize the move from cost-based to outcome-based AI deployment:

• An insurance carrier stopped counting claims adjusters and started counting settled claims. A major P&C insurer deployed an agentic claims processing system initially justified by a 30% reduction in handling time. Within 18 months, the frame shifted. The system was renegotiated on a per-claim-settled basis, with accuracy guarantees, and the insurer redeployed adjusters to complex claims that required human judgment. The cost savings funded a better customer experience, and the commercial model now rewards resolution rather than activity.
• A logistics provider priced on delivery promises kept, not trucks deployed. A global third-party logistics operator moved from per-shipment software fees to outcome-linked contracts, with pricing varying based on on-time delivery performance against committed SLAs. The AI system managing route optimization and carrier selection had a direct financial incentive to improve. Within two years, on-time delivery improved by 11 percentage points and the commercial model had become a source of competitive differentiation in customer negotiations.
• A financial services firm moved from cost-per-query to revenue-per-action. A wealth management platform deployed an AI advisory assistant originally scoped as a cost-reduction play against human advisor call volume. As the system matured, the commercial model shifted to a revenue-share arrangement where the platform’s fee was linked to assets under management, influenced by AI-recommended actions. The system’s incentive was now growth, not deflection.
• A healthcare system priced documentation AI on compliance outcomes. A large US health network deployed clinical documentation AI and moved from a per-physician license to a per-note pricing model contingent on regulatory compliance rates. The vendor’s revenue increased as accuracy improved. The health network’s administrative costs decreased as first-pass acceptance rates improved. Both parties had aligned incentives for the first time in the relationship.

The technologies and methodologies driving agentic value

The move from GenAI pilots to agentic value at scale requires a specific technology and methodology stack. It is not enough to deploy a large language model and call it AI transformation.

Agentic orchestration layers are the backbone of the shift. These are systems capable of decomposing complex, multi-step tasks into executable subtasks, assigning them to the right AI agents or human workers, monitoring execution, and escalating exceptions. Without orchestration, AI remains a point solution. With it, AI becomes a process engine.

Retrieval-augmented generation (RAG) and enterprise knowledge graphs ensure that AI systems operate on accurate, current enterprise data rather than hallucinating from training data alone. Outcome-based models require accuracy. Accuracy requires grounding AI in verified enterprise context.

Workflow automation platforms (including robotic process automation, integration layers, and API meshes) connect agentic AI to the systems of record where enterprise processes actually live. Intelligence without execution capability is just analysis. The combination of agentic reasoning and workflow execution is what makes outcome delivery possible.

Human-in-the-loop governance frameworks define when AI acts autonomously and when it defers to human judgment. Clarifying roles and responsibilities is more than a safety mechanism; it is a commercial and operational design principle. Outcome-based contracts require clear accountability, and that accountability requires a governance model that specifies the boundaries of autonomous action.

Intent-based interfaces are the emerging front end of agentic systems. Rather than configuring workflows through process diagrams or code, business users express outcomes in natural language. “Reduce days sales outstanding by 20% this quarter.” “Identify the top 500 accounts most likely to churn in the next 90 days and initiate retention workflows.” The system interprets the intent and orchestrates the execution, closing the loop between leadership mindset and operational reality.

Measurement and attribution infrastructure underpins outcome-based commercial models. You cannot price on business results if you cannot attribute those results to specific AI actions. Investment in AI observability, causal attribution, and outcome tracking is a prerequisite for the commercial models that will define the next generation of enterprise AI contracts.

The future of Services-as-Software: Where this goes next

The 4Ps framework is not a static snapshot. It is a direction of travel, and the direction is clear. More enterprises will make the leap from cost-and-scale to outcomes, and those that do it faster will compound the advantage.

Services-as-Software will evolve along three dimensions in the near term:

1. First, the unit of delivery will shift from output to impact. The market will mature past tokenomics toward full outcome accountability. Vendors who can credibly link their AI systems to measurable business results and who are willing to share the financial risk and reward of those results will win the largest contracts. Those who remain in consumption pricing will commoditize.
2. Second, the human role will shift from executor to arbiter. As agentic systems take over routine execution, the premium human contribution becomes judgment: strategic direction, ethical oversight, exception management, and relationship stewardship. The workforce transformation is not a replacement story. It is a reallocation story. Humans move up the 4Ps stack from productivity to performance.
3. Third, the boundary between software and services will continue to dissolve. The distinction between buying a software license and buying a managed service is already blurring. When an AI system executes a business process end-to-end, monitors its own performance, remediates its own errors, and bills based on outcomes delivered, the category label becomes irrelevant. What matters is the result.

The enterprises that will lead this transition are not waiting for the technology to mature. They are asking harder questions today: What outcomes do we actually want? What would we attempt if cost and scale were no longer constraints? How do we rewire our commercial models to reward the value we create rather than the activity we generate?

Those questions are the starting point for an intent-driven AI strategy. The data from 979 real-world use cases shows that the enterprises asking them are already moving up the 4Ps stack, from productivity to performance, from cost to outcomes, from pilot purgatory to production impact.

The Bottom Line: The Jevons paradox is coming for enterprise AI…

The only question is whether your organization is positioned to grow with it or get left managing a cost reduction that will soon be table stakes.