The New Economics of Selling AI Software

Software companies are moving beyond traditional subscriptions as AI introduces new costs et ways of charging customers. From usage-based fees to payments for completed tasks, the industry is experimenting with pricing models that reflect what AI actually does.

Artificial intelligence is changing how software companies charge their customers. Salesforce now offers AI agents priced by conversation or action, customer-service platform Intercom charges for completed outcomes, and European software group SAP uses credits that customers consume when its AI systems perform work. The changes are forcing software providers to reconsider the subscription model that has supported much of the industry’s growth.

For years, software-as-a-service (SaaS) companies have generally charged businesses according to the number of employees using a product. A company purchasing customer relationship management software might pay for 50 user accounts, irrespective of how frequently employees accessed the system.

AI complicates that arrangement because the cost of providing a service increasingly depends on how much work the software performs. An AI agent answering customer enquiries, searching company records or completing administrative tasks uses computing resources each time it operates. A customer with ten employees could generate more AI activity than another with 100.

Software providers must therefore find ways to recover their operating costs without giving customers unpredictable bills. Pricing also influences how businesses assess the technology, especially when AI is presented as an alternative to work previously performed by employees.

Software Companies Are Experimenting With Different Ways to Charge

Salesforce offers several pricing options for its Agentforce platform, including EUR 2 per conversation and consumption-based Flex Credits. Under the latter arrangement, customers purchase credits that are deducted when an AI agent performs actions such as updating customer records, answering enquiries or executing workflows.

Its European pricing lists 100’000 Flex Credits at EUR 500. A standard Agentforce action consumes 20 credits, which is equivalent to EUR 0.10 per action at that published rate.

A business cannot assume, however, that a completed task will cost only EUR 0.10. Resolving a customer problem may involve several actions, such as retrieving information, checking an account, updating a record and producing a response. Each billable action adds to the total.

Customer-service software company Intercom has adopted another approach. Its Fin AI Agent starts at USD 0.99 per outcome, with charges linked to defined results rather than the number of AI interactions required.

Fin can operate alongside an existing helpdesk without additional user-seat charges, or customers can purchase it through Intercom’s broader customer-service platform, which retains conventional per-user subscriptions.

The outcome model is intended to give businesses a clearer relationship between expenditure and completed work. Yet the definition of a billable outcome matters. Intercom includes certain completed workflows and handovers, not only customer enquiries that have been fully resolved without human involvement.

For buyers comparing software products, understanding these definitions is becoming as important as comparing the headline price.

European Software Providers Face the Same Calculation

SAP, one of Europe’s largest enterprise software companies, also charges for selected AI capabilities according to consumption.

Customers purchase SAP AI Units, which they can use across eligible premium AI services. In 2026, SAP prices its autonomous enterprise agents at 0.02 AI Units per action. An individual task may require several actions, depending on its complexity.

The system allows customers to allocate spending across different applications, although they must understand which activities consume credits and how their usage might change.

Swisscom provides an alternative example of AI pricing in Switzerland. Its Swiss AI Assistant, introduced in February 2026, offers a company-wide subscription with defined user and token allowances rather than charging separately for each action.

The Essential package costs CHF 149 per month, excluding VAT, and covers up to ten users with an allowance of 20 million tokens. Larger packages provide higher limits, with the Professional edition priced at CHF 399 per month for up to 40 users and 75 million tokens.

Swisscom markets predictable expenditure alongside Swiss data hosting, a consideration for companies handling confidential information or operating in regulated industries.

These offerings demonstrate that AI software is not converging on one pricing model. Providers are selecting different combinations of subscriptions, usage allowances and activity-based charges according to their products and the customers they serve.

Lower AI Computing Prices Do Not Necessarily Mean Lower Bills

The cost of processing individual AI requests has fallen for many models, but businesses may still spend more as employees use AI more frequently and applications perform increasingly complicated tasks.

An October 2026 contribution by technology consultancy EPAM Advisory to Handelsblatt Live examined the economics of token consumption, highlighting the importance of usage volume as well as the price charged for individual tokens.

AI models process text in units known as tokens. Providers may charge separately for the information sent to a model and the responses it generates, while additional costs can arise from document retrieval, external software calls and repeated attempts to complete a task.

For a SaaS company selling a conventional subscription with unlimited AI usage, heavier consumption can erode margins. Its customer may pay the same monthly fee while the cost of providing the service continues to rise.

Charging for usage protects the provider against some of that risk, but transfers greater uncertainty to the buyer. The commercial challenge is particularly acute for smaller software businesses that depend on third-party AI models and have limited control over their suppliers’ prices.

A provider may therefore choose to include a defined allowance in its subscription and charge separately once customers exceed it. Others may reserve more expensive models or agent functions for higher-priced packages.

Pricing decisions must account for the cost of running the technology, the number of customers likely to use it and the amount those customers are prepared to pay.

The Price of a Completed Task

Charging for a completed task can be attractive because corporate customers generally purchase software to achieve a result rather than consume computing resources.

A customer-service provider might charge for resolving an enquiry. An AI document-processing company could charge for each document successfully classified or processed. A sales platform might charge according to qualified enquiries, completed research or other contractually defined activities.

This approach makes the commercial proposition easier to explain, but suppliers have to establish precisely when work counts as complete.

An AI system might identify the correct document but produce an incomplete summary. A customer-service agent might answer an enquiry but fail to resolve the underlying problem. A workflow could perform several steps before requiring a human employee to finish the task.

In each case, the supplier and customer need a shared definition of success, supported by records that allow the result to be checked.

Outcome-based pricing also creates financial risk for software providers. A company promising a fixed price for each completed task may incur substantially higher computing costs when the work is unusually complex.

It must decide whether to absorb those costs, impose limits or charge separately for exceptional cases. Customers, meanwhile, need to understand how the product handles failed tasks, corrections and human intervention.

The agreement becomes particularly important when AI software performs activities involving financial transactions, customer records or regulated information.

Why Traditional SaaS Pricing Is Under Pressure

AI is also changing the relationship between software licences and employee numbers.

A business that uses AI to automate part of its administrative work may require fewer people to operate a particular system. Under a conventional per-seat subscription, that could reduce the software company’s revenue even as the customer relies more heavily on its technology.

A usage-based or task-based model gives providers another way to charge for the work their systems perform. It also allows customers to expand automation without purchasing additional licences for individual employees.

Traditional subscriptions are unlikely to disappear. Businesses still need software platforms, security controls, data management, administration and support. Many AI applications will continue to operate within established enterprise systems rather than replacing them.

Hybrid pricing consequently offers software providers considerable flexibility. Customers might pay a fixed subscription for platform access, receive an agreed AI allowance and incur additional charges according to consumption.

The appeal of such arrangements depends on their simplicity. Complex credit systems may suit technically experienced buyers, but finance and procurement departments still need to forecast expenditure and explain it to management.

What European AI Companies Need to Get Right

For European SaaS startups and established software providers, pricing has become part of product development rather than a decision made shortly before launch.

A company selling AI-powered financial analysis must understand how much each customer costs to serve, including model usage, data processing, infrastructure and support. It also needs to know whether customers value the number of analyses completed, the time saved or the quality of the results.

Those considerations influence which features belong in a subscription, which justify additional charges and how the product should be presented to prospective buyers.

European providers also have to account for local purchasing requirements. Swiss financial institutions may place particular weight on data hosting, confidentiality and contractual safeguards, while companies selling across the EU must explain their services within the applicable regulatory framework.

Customers buying AI software are increasingly likely to compare products on the basis of total expenditure rather than subscription prices alone. A low entry price may prove expensive if routine work generates substantial usage charges.

Vendors can make that comparison easier by publishing practical examples, explaining the units they charge for and providing tools that allow customers to estimate expenditure before committing to a contract.

As AI becomes part of everyday software, the number of users will tell suppliers less about the work their products perform. The companies that price that work clearly, and can demonstrate the results customers receive, will be better placed to defend both their margins and their commercial relationships.

Exporis Insight: How to Market AI Software When Pricing Is Changing

AI software companies need to explain their pricing as carefully as their technology. Corporate buyers want to know what a product will cost under normal working conditions, how charges are calculated and whether the proposed savings justify the expenditure.

Explain the billing unit. Identify whether customers pay per user, token, action, conversation or completed task. Where several charges apply, show how they combine on a typical invoice.

Use realistic customer examples. A demonstration of 100 successful transactions offers limited guidance to a company expecting 50’000 transactions a month. Pricing illustrations should reflect the volumes, complexity and operational requirements of the intended customers.

Show the full cost of adoption. Software companies should explain implementation, integration, training, support and additional usage fees, rather than concentrating exclusively on the monthly subscription.

Make performance claims verifiable. Suppliers claiming that AI reduces costs should explain their assumptions and distinguish projected savings from results measured in customer deployments.

Adapt communications to different buyers. Technical teams will examine integration and performance, while finance departments need forecasts and contract terms. Senior management will want evidence that the product improves business operations.

Exporis helps financial and technology companies communicate complex products through specialist content, product positioning and digital marketing. For AI software providers, explaining how pricing works can be an essential part of converting technical interest into paying customers.