Will AI Change the Economics of SaaS?

AI is making software easier to build while changing what it costs to operate, how customers pay for it, and where competitive advantage comes from. The question is not whether SaaS will disappear, but how its underlying economics will evolve when software begins to perform work rather than simply provide the tools to do it.

A Different Kind of Software Economics

For more than two decades, Software as a Service has been one of the most successful business models in technology. Customers pay recurring fees to access applications, while vendors spread the cost of developing and maintaining those applications across a growing customer base.

The model has attractive characteristics: recurring revenue, predictable customer relationships, and the potential for high margins as software scales. Once an application is built, serving additional users often costs relatively little compared with the value of their subscriptions.

Artificial intelligence is beginning to challenge several assumptions behind this model.

Generative AI can make software development more productive, potentially lowering the cost of building new applications. At the same time, AI-powered functionality introduces computational costs that increase with usage. And as AI systems move from assisting employees toward performing operational tasks, charging customers primarily for the number of people using an application may become less intuitive.

These developments have inspired competing predictions. Some expect AI to undermine traditional SaaS by making software easier to reproduce and replacing existing applications with intelligent agents. Others see AI as another technological capability that established vendors can integrate into their products.

Both perspectives contain elements of truth. Neither captures the full economic picture.

To understand what may change, it helps to examine four dimensions of the software business: what it costs to build, what it costs to run, how it generates revenue, and where it creates sustainable competitive advantage.

Software Becomes Easier to Build

Software development has traditionally required substantial investment in specialized technical expertise. Building an enterprise application involves designing architectures, developing functionality, integrating systems, testing, securing, and maintaining the resulting product.

Generative AI is changing parts of that process.

AI coding assistants can help developers write code, generate tests, explain unfamiliar systems, identify errors, and accelerate routine development tasks. More advanced tools can undertake sequences of development activities with varying degrees of human supervision.

Evidence of productivity improvements is emerging, although results differ considerably across contexts.

A Microsoft Research study combining three randomized field experiments involving 4,867 software developers found approximately 26% more completed tasks among developers with access to AI coding assistance.

Other findings have been less straightforward. Research organization METR reported in February 2026 that an earlier experiment had found experienced open-source developers working more slowly with AI tools. Its subsequent experiment faced methodological challenges that prevented a reliable estimate of current productivity effects.

The implication is not that AI development tools have little value. Rather, productivity gains depend on the nature of the work, the capabilities of the tools, and the environment in which they are used.

For new software companies, the potential is significant. Capabilities that once required months of engineering work may become easier to implement. Smaller teams can experiment with specialized applications, and products addressing relatively narrow business needs may become economically viable.

But producing working code is only part of delivering enterprise software.

A functioning prototype is not the same as a reliable application that hundreds of employees depend on every day. Security, regulatory compliance, system integration, data quality, performance, support, and long-term maintenance remain essential.

In complex business environments, much of the challenge lies not in writing code but in understanding what the software must actually do.

Business processes contain exceptions, dependencies, conflicting information, and rules that may be poorly documented or distributed across multiple systems and departments.

AI can support the implementation of those requirements. It does not automatically resolve their underlying complexity.

Lower development costs may reduce the barriers to creating software without reducing the barriers to delivering software that organizations can reliably operate.

This distinction becomes especially important when considering competitive threats to established enterprise applications.

But AI Changes What Software Costs to Run

Traditional SaaS economics depend partly on the ability to serve additional customers without proportionally increasing operating costs.

There are exceptions, of course. Infrastructure, support, data processing, and external service costs already vary with usage. Nevertheless, many conventional software operations can be performed repeatedly at relatively low incremental cost.

AI-intensive applications introduce a different cost profile.

Each interaction with a language model consumes computational resources. Complex tasks may involve multiple model calls, retrieval operations, tool executions, and verification steps. Agentic systems that monitor situations, interpret new information, and coordinate activities can generate ongoing costs as they perform more work.

These costs are not necessarily prohibitive. Nor should today's inference costs be treated as permanent. Model efficiency, specialized infrastructure, and improvements in execution strategies can substantially reduce the cost of individual operations.

But AI introduces a relationship between the amount and complexity of work performed and the cost of delivering the service.

For a software company, that raises important questions.

How much does it cost to complete a customer task? How often does the system need to retry an operation? What additional costs arise from monitoring, external integrations, or human review? And how does the cost of reliable execution compare with what customers are willing to pay?

The answers depend heavily on the application.

A simple classification task might require very little computation. A complex workflow involving many documents, multiple systems, and several rounds of reasoning can be substantially more expensive.

The relevant economic unit may therefore shift from the cost of supporting another user to the cost of completing another operational activity.

This does not mean AI-native software cannot achieve attractive margins. Companies can optimize model usage, combine AI with conventional software, reuse infrastructure, and reserve expensive reasoning capabilities for tasks that actually require them.

The important change is that vendors must understand the economics of AI execution, rather than assuming every additional unit of value can be delivered at negligible incremental cost.

What Happens to the Price per Seat?

The next question concerns how software companies charge their customers.

Traditional enterprise SaaS commonly combines subscription fees with pricing based on users, modules, storage, transactions, or other consumption measures.

Seat-based pricing is particularly intuitive when software serves as a tool for employees. If more people use an application, the company generally needs more licenses.

But what happens when software begins performing work that employees previously handled themselves?

Imagine an organization where a team of ten people manages a large volume of administrative processes. If AI-enabled software allows that team to handle twice as much work without adding employees, the customer's economic benefit increases even though the number of software users remains unchanged.

A pricing model based exclusively on seats may not reflect the additional value created.

If the organization eventually requires fewer people to perform the same work, the mismatch could become even more pronounced.

This creates incentives for software vendors to consider other approaches.

Usage-based pricing charges for measurable consumption, such as processed documents, executed transactions, or AI processing volume.

Workflow-based pricing relates charges more directly to completed operational activities, such as handling an incoming request or processing an order.

Outcome-based pricing connects payment to a defined result, such as a successfully resolved case or completed business process that meets agreed criteria.

Hybrid pricing combines a recurring platform fee with usage-based or performance-related elements.

These approaches are not entirely new. Enterprise software has long used transaction-based and volume-based pricing. AI may, however, increase their relevance as applications take on more operational work.

Bessemer Venture Partners examines these trade-offs in its AI Pricing and Monetization Playbook, highlighting the tension between predictable subscription revenue, the variable costs of AI execution, and pricing that reflects measurable customer value.

Each approach creates trade-offs.

Usage-based pricing can reflect activity levels, but customers may find variable costs difficult to predict. Workflow-based pricing can make value more tangible, but the underlying effort required to complete individual tasks may vary considerably.

Outcome-based pricing can align incentives particularly closely, but outcomes are often influenced by factors outside the software vendor's control.

A shipment arriving on time, for example, depends on carriers, ports, customs authorities, and other external parties. A software provider cannot reasonably guarantee every such outcome merely because it coordinates the process.

The more practical direction may be a combination of predictable subscriptions and charges tied to clearly defined units of automated work.

As software takes on more work, its economic value may become less closely tied to how many people use it.

A Different Business Model for Shipment Management

Consider a software platform used by a European importer to manage international shipments.

In a conventional SaaS model, the importer pays for access to shipment visibility, documentation, operational records, and collaboration features. Pricing may depend on the number of users, shipments, or tracked containers.

The platform provides a structured environment in which employees manage their logistics operations.

Its value comes from improving visibility, reducing information fragmentation, and making existing work more efficient.

Now imagine that AI-enabled capabilities allow the platform to assume part of that work.

The system might interpret incoming shipping documents, match them with relevant orders, request missing information from suppliers, update authorized business records, or coordinate responses to shipment delays.

The importer would still need reliable data, interfaces, reporting, and human oversight. But certain activities previously performed manually could now be carried out by software.

This creates a different economic proposition.

Suppose the importer manages 2,000 shipments annually.

If AI-enabled functionality could reliably eliminate an average of 15 minutes of manual effort per shipment, that would represent 500 hours of gross annual time savings.

At an illustrative labor cost of €40 per hour, the equivalent labor value would be €20,000 annually.

These figures are hypothetical, not evidence of realized savings. Actual benefits would depend on the activities automated, the remaining supervision effort, the reliability of execution, and whether released employee capacity could be productively redeployed or translated into lower costs.

Nevertheless, the example illustrates an important distinction.

The customer is no longer evaluating only the usefulness of software features. The customer can also evaluate the economic value of work the software performs.

For the vendor, this potentially creates room for pricing that reflects operational activity or value delivered.

But the vendor must also pay for AI execution, integrations, monitoring, and continued product development. If the cost of completing an automated task approaches or exceeds the price customers will pay for it, the business model will not be attractive.

The economics must work on both sides.

The customer's net operational benefit must justify the price, and the vendor's unit economics must support reliable delivery at scale.

That is a more demanding proposition than simply adding an AI assistant to an existing application and charging a premium for the feature.

If Software Is Easier to Build, Where Is the Moat?

Lower software development barriers raise an obvious competitive question.

If new companies can reproduce established functionality more quickly and cheaply, what protects existing software vendors?

The answer is not straightforward.

Established enterprise software providers often possess advantages that extend far beyond their codebases.

They may have thousands of customers, extensive integrations, years of accumulated business knowledge, mature security capabilities, and software deeply embedded in operational processes.

Replacing such systems can involve significant migration costs, business disruption, employee retraining, and organizational risk.

A new entrant must therefore do more than build a technically attractive alternative. It must provide a sufficiently compelling economic reason for customers to change.

At the same time, these incumbent advantages can create constraints.

Large applications often carry substantial architectural complexity. Existing customers depend on established functionality, interfaces, and workflows that cannot simply be discarded.

Changes to the underlying product architecture may involve considerable engineering effort and difficult migration decisions.

Commercial considerations matter as well. A vendor whose revenue depends heavily on user subscriptions may face uncomfortable trade-offs when introducing AI capabilities that reduce the number of users required to perform certain tasks.

This creates a tension between protecting an established business model and developing one that may eventually compete with it.

As McKinsey discusses in its analysis of software business models in the AI era, the shift toward AI-powered work is pushing vendors to reconsider not only pricing but also product design, customer relationships, and commercial operations.

This does not mean established vendors cannot adapt. Many have substantial resources, deep domain knowledge, strong distribution channels, and privileged positions within their customers' technology environments.

Nor does it mean new entrants are inherently better positioned.

The competitive outcome depends on how effectively each provider combines technological capabilities with the realities of enterprise operations.

AI-native newcomers may benefit from fewer architectural constraints and the freedom to design products around automated work from the outset.

Incumbents may benefit from trusted customer relationships, existing integrations, extensive domain expertise, and established channels through which new capabilities can be deployed.

Both groups face meaningful challenges.

More broadly, the basis of differentiation may shift.

When standard application functionality becomes easier to implement, its scarcity declines. Competitive advantage may increasingly depend on capabilities that are harder to reproduce: deep operational context, reliable integration with business systems, domain-specific execution logic, and the ability to perform consequential work under appropriate controls.

These advantages are not permanent or exclusive. They must be developed, maintained, and demonstrated through actual customer outcomes.

But they suggest that the value of enterprise software cannot be reduced to the cost of writing its code.

SaaS Is Not Disappearing. Its Value Proposition Is Changing.

Predictions about the death of SaaS often overlook what enterprise software actually provides.

Businesses need more than interfaces and application features. They need systems that structure data, maintain records, enforce business rules, manage permissions, support collaboration, and provide a reliable operational foundation.

AI does not eliminate these requirements.

In many cases, AI systems depend on them. An agent performing an operational task needs access to accurate business information, defined permissions, and systems capable of executing and recording transactions.

The result may therefore be an evolution of SaaS rather than its replacement.

Some applications will remain primarily tools used by people. Others will increasingly incorporate automated execution and coordination. Many will combine conventional interfaces, deterministic workflows, and AI-driven capabilities.

This creates opportunities for both existing vendors and new entrants.

It may also expand the software market.

If AI lowers the cost of developing and maintaining specialized applications, smaller industries and narrower business processes that previously could not support dedicated software products may become attractive markets.

At the same time, companies may become willing to pay more for software that delivers measurable operational benefits rather than merely supporting manual work.

The central economic opportunity is not that every application becomes an autonomous agent.

It is that software may capture a larger share of the value generated by completing business activities.

Whether this translates into higher revenues, better margins, or stronger customer relationships will depend on execution reliability, customer economics, competition, and the structure of individual markets.

The Economics of Getting Work Done

AI is unlikely to make the fundamental economics of enterprise software disappear.

Software will still need to be built, operated, maintained, secured, and sold. Customers will still evaluate costs, risks, and the value they receive.

But the balance between these elements may change considerably.

Development can become more efficient. Operating costs may become more closely connected to AI-driven activity. Pricing can evolve beyond traditional user licenses. And sustainable differentiation may depend increasingly on the ability to understand and execute real business processes.

These changes do not point toward one inevitable business model.

They point toward a broader reconsideration of what enterprise software delivers and how that value should be measured.

The future economics of SaaS may depend less on how many people use the software — and more on how much meaningful work the software helps complete.