How AI Is transforming SaaS pricing and shifting enterprise spending to labor budgets

For the last two decades, SaaS companies mostly competed for software budgets. Pricing was built around seats, modules, editions, and renewals because the human user was the best proxy for value. More users usually meant more workflows, more collaboration, more data inside the system, and more willingness to pay. AI changes that assumption.

When software can resolve a customer issue, process an invoice, qualify a lead, or review a contract with less human involvement, the seat becomes an incomplete proxy for value. The question is no longer just “how many employees need access?” It is “how much work can the system actually do?” That is the real pricing reset AI is bringing to SaaS.

The easy prediction is that AI will kill seat-based pricing. That may not be accurate; seats will remain important because enterprises still need access control, permissions, security, reporting, governance, and auditability. But seats will become only part of monetisation; the bigger opportunity is that enterprise software can move from pricing access to pricing work. Traditional SaaS sold into the software budget. AI opens a larger budget pool: labor. If an AI agent can reduce support volume, automate invoice processing, improve collections, qualify leads, or reduce back-office effort, the buyer is no longer comparing the product only to another software vendor. They are comparing it to headcount, outsourcing, shared services, consulting, and operational inefficiency.


That is a much bigger opportunity. It is also a much bigger risk.

A vendor that reduces human effort but still charges only by human seats may create more value while capturing less of it. The companies that win will not simply be the ones that add AI features but the ones that can prove they own a meaningful unit of work and price accordingly. This is where token-based pricing comes into play, it’s a transitional step but not the final answer because tokens, model calls, prompts, and compute matter are a vendor cost unit, not a customer value unit. No CFO wants to approve a business case around tokens as they care about completed work - was the support issue resolved? Was the invoice processed? Was the contract reviewed? Was the payment collected? That is the language of business value.

The most elegant pricing model is outcome-based pricing: the software completes a valuable task, and the customer pays when the outcome is achieved. It’s harder to implement as it only works when the outcome is measurable, attributable, auditable, and easy to trust. Let’s take customer support as an example - if an AI agent resolves a support issue without escalation, there is a recognisable unit of value. But many enterprise workflows are messier. What is the exact value of better forecasting? How much legal risk was avoided by reviewing a contract? What portion of a sales conversion was caused by the AI agent? If the customer does not trust the measurement, they will not trust the invoice. That is why the future is not simply seats, usage, or outcomes. It is a hybrid stack: access, execution, and outcomes.

LAYER	WHAT THE CUSTOMER PAYS FOR	EXAMPLE
Access	Right to use the platform	Seats, Roles, Editions
Execution	Work performed by AI	Credits, Actions, Documents processed
Outcomes	Verified business results	Resolved cases, Collected payments, Qualified leads

The strategic question is who captures the pricing power. Incumbents like Salesforce, SAP, ServiceNow, Workday, Microsoft, and leading industry software platforms have an advantage because they already sit close to enterprise workflows, hold data, permissions, integrations, trust, and distribution. But that advantage is not guaranteed as owning the system of record does not automatically mean owning the work.

A company may own customer data in a CRM, employee data in an HR system, financial data in an ERP, or contracts in a CLM system. But the actual AI work may be performed by another layer sitting above those systems. That layer could be a frontier AI company like OpenAI or Anthropic, a systems integrator like Accenture or Deloitte, an internal AI team, or an AI-native company like Sierra.

PLAYER	WHY THEY CAN WIN	WHAT COULD GO WRONG
Systems of record - Salesforce, SAP, Workday, ServiceNow, vertical platforms	They already own trusted workflows, data, permissions, and enterprise context. They win if they become systems of action.	They may keep the database while another AI layer captures the work and the labor budget.
AI-native work layers - Sierra, agentic startups	They can sit above existing systems and own a completed task, such as customer resolution or sales follow-up.	They need deep integrations, customer trust, and clear outcome measurement.
Model providers - OpenAI, Anthropic, Google	They can become the interface where enterprise work begins, not just the intelligence layer behind someone else’s product.	They may be abstracted by applications, platforms, or services firms.
Services firms - Accenture, Deloitte, PwC, Wipro	They can capture the messy middle: workflow redesign, integration, governance, and change management.	They may capture spend, but not always durable product-like revenue.
Systems of store - passive CLM, document repositories, knowledge bases	They can win only if they evolve from storing information to owning workflow or action.	They are most exposed because AI can summarise and search information from outside the system.

The real threat to SaaS incumbents is that they may keep the database while someone else captures the labor budget. Sticky software will get time. Systems of record will get leverage. Systems of action will get pricing power. The next generation of pricing will not be defined by seats versus tokens but by who owns the work, who does the work and who gets paid for it.