Imagine a software vendor adds an AI feature that allows a team of 40 people to do work that previously required 400.
The customer is getting considerably more from the product. However, if that product is licensed per seat, they may need fewer licenses when they renew. The vendor has improved the product and increased its value while reducing the number of licenses the customer needs.
This is starting to expose a weakness in a model the software industry has relied on for decades. AI is changing the relationship between headcount and software use while introducing new costs for the vendor each time AI is used. The latest IDC research commissioned by Thales also shows growing buyer interest in usage-based models.
Moving away from per-seat pricing, however, depends on whether vendors can measure and manage whatever they decide to charge for instead.
Why Per-Seat Is Failing Now
Per-seat licensing made sense when the number of people using a product was reasonably closely tied to the value it delivered. A growing company hired more employees, bought more licenses, and paid the vendor more.
AI changes those economics because output can increase without increasing headcount. An AI agent may perform tasks that previously required another employee, while AI features can increase the amount of work existing employees complete.
Gartner is already seeing this in enterprise AI coding agents, where pricing is shifting from seat-based licenses toward hybrid and usage-based structures. CIO has reported on the implications for SaaS more broadly, citing Gartner estimates that agentic AI could put $234 billion in application software spending at risk by 2030 as agents increasingly become users of business applications.
Software vendors are responding. IDC figures supplied for this research show that 76% say AI is prompting them to consider new pricing models, with 58% identifying usage-based pricing as their leading approach.
The pressure is also coming from customers. Software publishers are hearing from customers who want agentic workflows and, as a result, fewer seat licenses.
That does not necessarily mean customers are resistant to the change. If AI removes a task, that is one less thing for someone to do. There may be a natural question about what that means for an individual’s role, but in many cases, the role evolves as AI takes over more of the work.
There is also the cost of providing AI functionality to consider. Inference, compute, model calls, and data processing create costs that rise with consumption. If a vendor continues to charge a fixed amount per user, it can end up serving fewer licensed users who consume more expensive AI functionality.
A report by Thales and Simon-Kucher identifies the same problem: traditional user- or device-based licensing may no longer reflect how customers derive value from software, particularly as AI reduces the number of users required.
Why Moving Away From Seats Is Difficult
Deciding that a product needs a different pricing model is often much easier than putting that model into production.
Publishers are not necessarily waiting for renewal cycles to make that decision. Some are already exploring specific use cases for moving away from seat-based licensing as the subscription economy begins to give way to a usage economy.
IDC figures show that only 25% of vendors have centralized usage data. Legacy integration is the leading barrier for 71%, while 61% still manage usage through in-house tools or Excel. Only 10% have fully automated the process.
This comes up regularly in conversations with software companies. A business decides to sell an AI capability using credits or tokens and assumes its existing billing system will handle the change. Then the implementation starts.
Many billing systems were designed to issue a fixed subscription invoice every month or year. They may not be able to meter several types of consumption, reset token allowances, reconcile usage against entitlements, or account for consumption coming from different deployment environments.
Companies can get surprisingly far into a project before discovering those limitations. By that point, the pricing model may have been agreed upon, the product may be close to launch, and the operational systems may be trying to catch up.
The Thales and Simon-Kucher report describes the problem: “Identifying the right monetization model is a strategic decision. Executing it is an operational one. The distance between those two things is where revenue is won or lost.”
What Has To Happen Behind Usage-Based Pricing
Usage may come from a SaaS application, an API, an AI model, a connected device, or software running on the customer’s own infrastructure. Vendors need to collect that information wherever the product is deployed, including environments where cloud telemetry is unavailable.
That does not mean vendors can re-engineer every product around a new usage model. Most do not have that luxury. The more practical approach is to work with the usage data the product already produces, ingesting it in its existing format rather than requiring a major rebuild.
That reduces the engineering work involved and allows the move to usage-based pricing to be treated more as a go-to-market or new-product motion than a product re-engineering exercise.
Those raw usage events need to be converted into records that support billing. Finance needs reliable data before it can invoice against consumption, and customers need to be able to reconcile a charge with what they actually used.
Entitlements add another element to the equation. If a customer has bought one million tokens, the vendor needs a current view of how much has been consumed and how much remains. It also needs to know what happens when the customer approaches or exceeds that allowance.
What happens when a customer reaches their allowance will depend on the contract. The vendor might issue a warning at 80%, stop consumption at the agreed limit, allow an overage, or offer additional capacity.
The billing system does not necessarily have to perform all of this work. Usage can be collected, processed, and reconciled against entitlements before the resulting information is passed into existing invoicing and revenue recognition systems.
Flexible licensing models can then sit alongside the finance systems a company already uses.
Most Vendors Will Need More Than One Model
For established software companies, a mix of models will matter more than a wholesale move from seats to consumption. Different business units often sell in very different ways. One may use annual subscriptions, while another needs monthly terms that allow customers to scale up and down. A common monetization platform has to accommodate those differences.
Hybrid pricing is one way companies are doing this. A product might retain a base subscription with an included usage allowance, charge for consumption above that level, and give customers the option to buy additional credits. The research identifies hybrid models as a practical way to introduce consumption-based pricing while retaining a predictable base level of revenue.
Per-seat pricing will continue to work where the number of users remains a reasonable measure of customer usage. AI-enabled products are making that calculation less straightforward.
Software companies may find themselves running several commercial models simultaneously. Some products will remain seat-based, others will use credits or consumption, and some will combine the two.
Supporting that mix means being able to consistently collect and manage usage across products without having to build a separate process every time the commercial model changes.

