SAP overhauls its AI strategy: new teams and pay-per-use pricing instead of subscriptions

SAP is making one of the most visible reversals in its product strategy in years. CEO Christian Klein is assembling a new AI unit of several hundred employees and preparing a shift from classic subscriptions to a model in which customers pay for their actual use of AI features.

Why change the model

For SAP, this is not a cosmetic price-list update but an attempt to reshape the whole business around the market's new logic. Historically, enterprise software was sold by the number of users, seats or modules. But when AI assistants and agents start performing routine operations on their own, that scheme no longer reflects the product's value accurately. A single well-integrated AI scenario can automate the work of several employees, and in that situation charging simply for access to the system becomes less and less convenient and less and less convincing for SAP.

The company is under pressure from two sides. Externally, competition is growing from generative AI platforms that promise fast results without heavy implementations. Internally, SAP has to speed up its own restructuring: in early March, Klein already redistributed management responsibilities to focus more on product and artificial intelligence. The next step is launching dedicated engineering teams from July, which will work closer to customers and help bring AI scenarios to real-world use.

What customers will get

The new model implies not only a different price list but also a different way of deploying software. SAP intends to create so-called forward-deployed engineering teams: mixed teams of consultants and developers who will work with the customer to build applied AI solutions for specific processes. This is not a "magic button" on top of ERP, but the embedding of AI into finance, procurement, supply chains, HR and other areas where context, data and control matter.

• Dedicated teams to co-develop AI scenarios with the customer
• More customized solutions for specific processes, not just standard features
• A shift to payment for actual AI use, not only for the number of users
• A bet on AI assistants and agents embedded in existing business systems

For customers, this may prove more convenient than a familiar subscription, especially if AI use is uneven or starts in just a few processes. In its own materials, SAP already shows it is moving toward a usage-driven model with AI units and access to Business AI through Joule. The idea is simple: if AI speeds up the quarterly close, automates the preparation of commercial proposals or removes some manual work, the price should be tied to the volume and effect of use, not to an abstract number of licenses.

Where the main risk lies

This reversal is painful for SAP precisely because it touches the foundation of the old enterprise software economics. A subscription provides predictable revenue, while a consumption model makes income more variable and ties it more closely to proven results. On top of that, the company still carries the legacy of its previous transition, from on-premises software to the cloud, which took years and remains a sensitive topic for customers, partners and investors.

There is a second risk: deploying AI in a real business workflow is far harder than showing a polished demo. SAP itself wrote at the end of March that AI is already part of more than two-thirds of its recent cloud deals, but that does not yet mean mass industrial impact. For an agent to work in finance or the supply chain, it needs quality data, approval rules, audit, access control and predictable processes. If SAP cannot quickly prove the value of the new scheme, some customers may keep their core on SAP but build the AI layer with external models and contractors.

What this means

SAP is effectively acknowledging that AI changes not only the interface of enterprise software but also how it is monetized. If Christian Klein's experiment works, the market will get a strong signal: in the era of AI agents, selling enterprise products only by user count is no longer enough, and the winners will be vendors that can tie AI to a concrete business result.