AI Is Changing the SaaS Model: Pay for Work Done, Not Seats Filled

The per-user SaaS model was designed for a world in which humans operated software.

Every employee needed an account. Every account represented access to a set of features. As a company hired more people and adopted more applications, its software bill grew with its headcount.

That model was simple to understand, easy to sell and highly predictable for software vendors. It also helped replace the large upfront licence agreements that dominated enterprise software for decades.

But artificial intelligence is beginning to break the relationship between the number of users and the value created by software.

When an AI agent can configure a system, prepare reports, create quotations, update opportunities, reconcile information or automate a workflow, the seat is no longer the only actor performing work.

That leads to a simple but uncomfortable question:

Odoo is an example, not the problem

The recent debate around Odoo Enterprise pricing has made this tension particularly visible within the Odoo ecosystem.

Odoo remains one of the strongest value propositions in business software. It brings a broad collection of integrated applications into a single platform and remains substantially more accessible than many traditional enterprise systems.

This article is not an argument that Odoo should not set its own prices. Every software company must fund development, infrastructure, security, support and long-term product evolution.

The interesting point is the unit used to calculate that value.

Odoo's current Enterprise model is still largely based on the number of users, light users and the selected subscription plan. At the same time, its Custom plan now includes Agentic AI. That combination is a useful illustration of a much wider industry transition: software is beginning to perform work, but many commercial models still charge primarily for the humans accessing it.

This is not uniquely an Odoo issue. It affects almost every SaaS company whose economics were built around seats.

A practical perspective from the Odoo ecosystem

José Antonio Ibarra Ossandón recently published a useful analysis of Odoo's pricing changes in North America and three ways companies can reduce their licensing costs. His video also explores the role of Odoo Community-based alternatives and mentions Binhex Cloud as one example of how that ecosystem can be delivered as a managed platform.

The video is in Spanish, but it provides valuable context for partners and customers evaluating the long-term cost of their Odoo strategy:

Watch “Odoo increased prices by 30%: 3 ways to reduce licensing costs” on YouTube

The seat was a proxy for value

Per-user pricing was never a perfect measurement of value. It was a convenient proxy.

If 100 people used a CRM, the assumption was that the company obtained more value than a business with ten users. In many cases, that was directionally correct. More users generally meant more activity, more records, more collaboration and more reliance on the platform.

However, the model also created distortions:

• Companies restricted access to control costs.
• Occasional users were kept outside the system.
• Teams shared information through spreadsheets, email or messaging tools instead of giving everyone access to the source of truth.
• Adoption became financially penalised: the more successfully a company involved its employees, the larger its recurring licence bill became.
• A user who opened the system twice a month could cost the same as one who worked in it every day.

The arrival of light-user licences attempts to reduce some of these problems. But it does not change the underlying assumption that the person accessing the system is the primary unit of value.

AI challenges that assumption.

AI becomes another operator of the business

Traditional software waits for a user to decide what to do, navigate to the correct screen, enter the information and confirm the operation.

Agentic software can work differently.

A user can describe an objective in natural language, and the system can determine which applications, records and actions are required. With the appropriate permissions and confirmation controls, an AI agent can perform several operations across the company platform.

The user may ask:

• Which opportunities require attention this week?
• Prepare quotations for the customers we discussed.
• Create follow-up activities for every overdue proposal.
• Configure sales tax for the states where the company has nexus.
• Import this document and update the corresponding records.
• Compare this month's sales with the previous quarter and explain the variation.

The value is no longer created only when a person clicks through an interface. It is also created by the work the agent performs on the person's behalf.

One agent may execute thousands of operations across sales, accounting, inventory, purchasing or customer service. Meanwhile, dozens of employees may need occasional access to review information, approve decisions or collaborate in a process.

In that environment, headcount becomes an increasingly weak representation of software consumption.

AI should be the interface, not another application

Many software products currently treat AI as one more feature: a chat window placed next to the existing interface.

That can be useful, but it does not fundamentally change the operating model.

The larger opportunity is to make AI the primary interface between people and the company's systems.

The employee expresses an intention. The AI understands the context, selects the appropriate tools, proposes or executes the necessary actions and keeps the user in control of consequential decisions.

Underneath that interface, the company still needs a serious transactional platform. AI does not eliminate the need for:

• Structured accounting records.
• Inventory movements and valuation.
• Sales, purchasing and manufacturing workflows.
• Permissions, auditability and company rules.
• Localisations and regulatory compliance.
• Reliable integrations and data models.

The interface may become conversational and agentic, but the foundation must remain robust.

Why open source becomes more important in the AI era

An AI agent becomes more useful as it gains access to more of the company's operations. That also increases the importance of control, transparency and portability.

Businesses should be able to understand where their data is stored, which actions an agent can perform, which models or providers are used and how they can change those components over time.

An open-source foundation offers an important advantage: the operating layer of the company does not have to belong entirely to a single vendor.

The Odoo ecosystem provides a concrete example.

Odoo Community supplies an integrated open-source business platform. The Odoo Community Association maintains a large collaborative ecosystem of repositories and modules extending areas such as accounting, logistics, reporting, subscriptions, connectors, localisations and technical infrastructure.

Historically, the weakness of this alternative was not necessarily a lack of capability. It was complexity.

Selecting modules, resolving dependencies, hosting the platform, maintaining environments, applying updates and managing upgrades required considerable technical knowledge. That made Odoo Community and OCA powerful, but less accessible to companies seeking a packaged experience.

AI and managed infrastructure can close that gap.

AI can guide configuration, install applications, automate workflows and operate the system. A managed platform can provide deployment, backups, monitoring, staging, upgrades and security. Odoo Community and OCA remain the open foundation underneath.

The result is not simply a cheaper ERP. It is a different software architecture:

Pay for work done, not seats filled

If AI becomes an operator, then usage becomes a more relevant unit of value than access alone.

A future commercial model could combine:

• A platform subscription covering infrastructure, security and maintenance.
• Capacity for storage, workers and service levels.
• AI consumption based on the resources or operations used.
• Optional implementation, support and specialist services.
• Additional connectors or regulated services where external costs exist.

In this model, adding an employee does not automatically add another full recurring licence. More people can participate in the company's processes without adoption becoming a financial penalty.

The company pays more when it consumes more infrastructure, requests more support or asks AI to perform more work—not simply because another person needs visibility into the business.

This aligns price more closely with activity and value.

Predictability does not require seat-based pricing

The strongest argument for per-user pricing is not that seats measure value particularly well. It is that they make expenditure relatively easy to forecast.

Work-based pricing sounds fair until finance has to budget for it. A model that charges separately for every action could create uncertainty, discourage adoption and replace a familiar licence bill with an equally problematic meter.

But predictable pricing does not have to remain tied to headcount.

The market may follow a path similar to telecommunications. Customers once paid for every call or message. Over time, providers moved towards flat-rate plans built around expected usage and network capacity. The individual action stopped being the principal billing event, even though the underlying infrastructure still had a cost.

AI pricing could develop in a similar direction: a predictable platform fee, an included amount of AI capacity, transparent tiers, spending caps and planned upgrades when additional capacity is required.

As models become more efficient—and open source enables organisations to run private or dedicated LLM infrastructure—the underlying economics may increasingly resemble reserved compute capacity rather than a charge for every interaction. Hardware, energy, operations and peak demand still matter, but those costs can be packaged into understandable capacity bands.

The alternative is therefore not seat-based pricing versus unpredictable usage. It is headcount-based pricing versus predictable capacity-based pricing.

Usage-based pricing is not automatically better

Changing the unit of value does not solve every problem.

Poorly designed usage pricing can become unpredictable. Customers may hesitate to use the product because they fear an unexpected bill. Vendors may hide complex calculations behind credits that are difficult to understand. And not every AI operation creates the same value.

A responsible model therefore needs:

• Clear definitions of what is being consumed.
• Real-time visibility into usage.
• Predictable allowances and spending limits.
• Controls that prevent accidental or abusive consumption.
• Pricing that reflects meaningful work rather than arbitrary technical events.
• The ability to scale down as well as up.

The goal should not be to replace one opaque pricing model with another.

It should be to align the economics of the platform with the value it produces.

Software is not becoming free

Open source does not mean zero cost, and AI does not remove the need for experts.

Business software still requires infrastructure, implementation, migration, training, support, maintenance, security and continuous improvement. Complex organisations will continue to need consultants who understand their processes and can take responsibility for important decisions.

What changes is where those resources are spent.

Instead of paying primarily for permission to access the software, companies can invest in outcomes: better implementation, stronger support, more automation, safer infrastructure and more work performed by AI.

That is a healthier conversation than simply asking how many licences the company needs.

From SaaS to the Agentic Enterprise Cloud

The SaaS model moved business software from servers owned by each customer to applications delivered through the cloud.

The next transition may move us from applications operated entirely by humans to platforms operated jointly by people and AI agents.

That is the idea behind an Agentic Enterprise Cloud:

• AI becomes the primary interface.
• A complete business system provides the operational foundation.
• Humans retain permissions, oversight and accountability.
• Open standards preserve flexibility and portability.
• Pricing follows capacity and work performed rather than headcount alone.

This does not necessarily mean the end of SaaS.

Subscriptions will remain useful. Cloud infrastructure will remain essential. Vendors will still need recurring revenue to build reliable products.

But the user licence may no longer be the best primary unit of value.

The companies that recognise this shift early will not merely add AI to their existing applications. They will redesign how users interact with software, how the software performs work and how customers pay for the value created.

The future of business software may be simpler than the industry expects:

Further reading

• Odoo pricing
• Odoo Enterprise Subscription Agreement
• Odoo Community Association repositories
• Odoo Community Association
• Odoo increased prices by 30%: 3 ways to reduce licensing costs — José Antonio Ibarra Ossandón