Evaluating the Commercial Models for Agents

When discussing agent commercialization, the key concept to understand is agent steps. Steps are the basic unit used to measure and price the work an agent performs.

An agent step is any action taken by the agent while solving a request — for example, planning the next move, making a tool call, or running a reasoning cycle. In short, every meaningful "piece of work" an agent does is counted as a step.

Steps are used because they scale naturally with both value and cost. A simple request might require only a few steps, while a more complex one could take many more. This ensures that customers pay fairly: low-effort tasks are inexpensive, while more complex interactions are charged appropriately.

The model also provides consistency across agent categories. Whether the agent is SAP-built or customer-built, the same logic applies. Customers can clearly see how usage is measured, and pricing aligns with the value delivered.

Pricing is then tied to the agent category. Basic, standard, and advanced agents each have their own per-step cost, and steps always match the category. A basic agent only executes basic steps, while an advanced agent performs advanced steps. The idea is like employee roles: a junior and a senior may complete the same task, but the cost differs because the expertise level is different. Advanced agents use more powerful models and tools, so each step costs more — but they may also complete the task in fewer steps.

This step-based approach is not unique to SAP. It is becoming an industry standard, with other providers like Salesforce and ServiceNow adopting similar models under different names, such as "actions." In practice, most agent tasks fall into a range of 2 to 20 steps, depending on complexity.

In summary, Agent steps offer a fair, transparent, and consistent way to commercialize agents, directly linking costs to the actual work performed.

In the previous section, we established that agent steps are the core unit for agent commercialization. Now we’re looking at how those steps translate into AI Unit consumption, depending on the commercial model.

For PUPM packages, agents consume requests from the package, and those requests are consumed through steps. Each step an agent executes draws from the user’s request allowance.

Every user has a fair-use pool of requests, and unused capacity can be shared across users within the same package. Overage only applies once the shared pool is fully consumed.

For agents outside of PUPM packages, steps are monetized directly through AI Units.

This applies to:

• SAP-built agents that are not offered as part of a PUPM package, such as Signavio or LeanIX
• Customer-built agents created with Joule Studio
• Indirect access scenarios (A2A)

In all cases, consumption is driven by agent steps.

The cost per step is determined by the agent’s category — Basic, Standard, or Advanced.

• For SAP-built agents, the category reflects the business value delivered by the agent.
• For customer-built agents, agents today generally align with the Basic category, with the potential for further differentiation over time as capabilities evolve.

The important point is that steps always match the agent’s category. A Basic agent performs Basic steps, and an Advanced agent performs Advanced steps — there is no mixing. Higher-category agents have a higher cost per step, but they often complete tasks more efficiently.

So, whether usage is covered through a PUPM package or billed consumption-based, every agent step is consistently measured, and pricing scales with the value and complexity of the work performed.