Pricing agentic AI: Working toward a framework for value and ROI

As agentic AI makes the leap from a theoretical concept to an enterprise reality, the question of how to price it has taken center stage. Traditional pricing models, such as straightforward seat-based licenses of the SaaS era or the metered usage of the cloud, are falling short. The autonomous, value-driven nature of AI agents requires a more sophisticated pricing strategy, one that better defines, and captures, the value they deliver while clearly tying their efforts to return on investment (ROI) for enterprises.

Startups and their efforts at enterprise engagement on agentic AI are critical in this evolving conversation. As they respond to ever-evolving market dynamics, startups are iterating on three primary models, each with its own unique value and risk attributes. We will examine those three pricing models and also hear from startup founders on what is working, what is not and what to expect in the day, months and years ahead.

Outcome-based pricing

The most direct agentic AI pricing structure is outcome-based, a model that lays out a clear and compelling ROI story. Here, enterprise costs are not tied to usage or access, but instead to a specific, measurable business outcome: customers only pay for the tangible results the agentic AI produces. In the customer experience space, for example, a company might pay only for each support ticket an AI agent successfully resolves without human help.

Here, the ROI is immediately apparent through reduced labor costs, as every ticket handled by an agent is categorized as a direct saving. This model also drives efficiency, with agents working 24/7 to increase resolution volume and speed. Importantly, this model guarantees performance by shifting the financial risk to the vendor – and enterprise pays only for success, ensuring every dollar spent generates a positive outcome.

This paradigm has proven highly effective for startups whose growth is directly linked to their customers' financial and operational goals.

Nikola Mrkšić, CEO and Co-Founder of PolyAI, said regardless of the pricing model that a startup may be using, they need to tailor their pricing models to the specific needs of potential buyers.

“Startups have to go and build a financial model that is really from the customer’s perspective, and people don’t always want to share all that data with you. You need to have very compelling modelling of their cost basis to know how to price,” Mrkšić said.

Consumption-based pricing

With consumption-based pricing, ROI is framed through the lens of operational efficiency. In this pay-per-use model, cost is linked to the computational resources consumed, such as the number of API calls made or the volume of tokens processed.

This approach offers tremendous flexibility, allowing companies to scale their use of AI without a large upfront investment. The return is realized as teams automate a wide range of tasks, from generating marketing copy to analyzing data, paying only for what they use. This frees up valuable employee time for higher-value strategic work.

The low initial cost encourages broad experimentation, allowing departments to discover new efficiency gains over time. From the vendor's side, this model aligns their own costs with revenue. For the customer, however, the challenge lies in managing predictability, as highly variable agent usage can lead to unexpected expenses. The ROI is clear, but it requires careful monitoring to ensure costs don't outpace the value gained.

Enterprises who see success with this model but want more predictability on future expenses are already angling on future cost control options.

“Everyone is trying to think two steps ahead. We have customers who are explicitly asking us to bring their entire agentic execution into local models on local hardware. No pricing discussion is complete without the notion of control and independence and choice. People don’t want to be beholden to anyone,” said Avinash Misra, CEO and Co Founder of Skan AI.

“At the back of that calculus is the build up toward ‘How will costs be controlled?’ and ‘How will they be paid for?’”

Seat-based pricing

Last, the familiar seat-based pricing is being adapted for the agentic age, casting it as a productivity multiplier. In this traditional SaaS model, a company pays a recurring fee for each employee licensed to use specific software. While straightforward on paper, this model's ROI is rooted in its ability to amplify a workforce’s capabilities. For example, an agent can help a single employee perform tasks that previously required an entire team. An in-house counsel, for instance, could use a legal agent to review contracts in a fraction of the time, dramatically increasing their output and capacity. For the enterprise CIO, the fixed cost per user offers the comfort of simple, predictable budgeting and encourages the widespread adoption of the tool. The primary risk, however, falls on the vendor, who may severely undervalue their technology. If a single agent provides a tenfold productivity boost or effectively replaces the need for additional hires, a simple seat license fails to capture that exponential value.

Mrkšić said it can be difficult for companies to compare human and bot activities, especially when it comes to outcomes and, thus, agentic pricing.

“Every company looks at ROI calculations differently. The main thing we look for when discussing pricing is metrics: do companies know how to measure what their people are doing versus what software or, now, agents, will be doing,” he said.

Conclusion

Ultimately, there is no single, perfect answer to pricing agentic AI. The optimal model is a nuanced choice that depends on the specific use case, the measurability of the outcome and the risk tolerance of both the buyer and the seller. The fundamental task for any company in this space is to construct a pricing framework that transparently connects its cost to the transformative ROI it can deliver – whether that value is found in direct cost savings, scalable efficiencies or an exponential increase in human productivity.

“Any pricing model requires a deep understanding of the work being done and the role of people in that work. Even on the SaaS side, where human labor is involved. Companies know the most about their people and their costs. So when we are talking about agentic pricing, it helps for them to bring these insights to the table, since the role of people is changing,” Misra said.

“This is what is under stress. These two worlds are being disturbed, because the fundamental unit of software usage, which was the human getting productivity from software, is now part of a larger conversation that includes autonomous agents. Therefore, what is the price people will pay, and this whole shift in the theater from human execution to agent execution – that’s what is causing the generational transition in value creation and value capture.”