SAP AI Units can feel confusing at first.
And honestly, that is understandable.
SAP is trying to create a simpler way to package and charge for premium AI capabilities across its portfolio. But if you are an IT leader, a consultant, or someone who has to explain the monthly bill, the question is much more practical:
How do we make sure the bill does not get out of control?
That is what we are going to dive into in this article.
We are going to break down, in plain English, how SAP AI Units work, how Joule and AI agents may consume them, where costs can start to grow, and what precautions you should take before a big surprise lands in your inbox at the end of the month, when the damage may already be done.
The hard work happens earlier.
Before you start experimenting and let AI run wild across teams, you need to understand a few things first:
- How consumption is generated.
- How to set limits.
- Who owns the spend.
- How to design agents carefully to optimize AI Units.
- And how to choose the right use case in the first place.
Before getting into the practical side, it is worth taking one step back. In a previous article, we looked at the broader SAP tokenomics debate: why SAP is trying to move the AI pricing conversation away from raw token usage and toward business outcomes, and why AI Units are becoming part of that shift.
If you want the broader landscape first, you can read it here: What Is the General Landscape of Tokenomics in SAP?
Now, to the matter at hand: let’s look at the details of cost inside SAP AI offerings.
Let’s Start with the Basics: What Are SAP AI Units?
Think of AI Units as SAP’s commercial currency for consuming certain premium SAP Business AI capabilities.
They are not exactly the same as technical AI tokens. A token is a lower-level unit used by AI models to process text. An AI Unit is SAP’s business-facing way to package, meter, and charge for premium AI usage across its portfolio.
SAP Business AI uses a hybrid commercial model that combines per-user access for premium AI capabilities with AI Units as a virtual currency. Customers purchase AI Units annually and draw from a central pool that can be used across different SAP products and AI services.
In simple terms:
Base AI is included with your SAP product licensing. These are the more standard Joule capabilities, such as navigational, informational, transactional, and simple analytical features. Premium AI requires AI Units. This includes more advanced capabilities, embedded AI features, and agent actions executed by Joule Agents.
So the first control question is very basic:
Are we using something included, or are we consuming AI Units?
How AI Units Are Measured
Here is where it can get a little messy:
AI Units are not always consumed in the same way. Depending on the SAP capability, consumption can be tied to users, requests, products, features, or agent steps.
For example, SAP Joule for Consultants uses a per-user-per-month model. SAP Learning says it charges 35 AI Units per assigned user per month, includes 22,900 requests per assigned user per month, pools requests across assigned users, and charges overages at 2 AI Units per 1,000 requests. Billing is based on the highest number of assigned users during the month, also called the high-watermark model.
That means one small operational detail can matter a lot.
If you assign 100 consultants to Joule for Consultants, the spend is not only about how much each person uses it. It is also about who has access, how many users were assigned at the peak of the month, and whether total usage exceeds the included request pool.
Now look at agents.
SAP Learning explains that agent usage is measured in steps. In consumption-based scenarios, Basic Agents consume 0.005 AI Units per step, Standard Agents consume 0.01 AI Units per step, and Advanced Agents consume 0.025 AI Units per step.
That may sound tiny.
But agents multiply tiny numbers.
Imagine a basic agent that checks a supplier exception. It reads the case, checks vendor data, validates a policy, drafts a response, and routes the issue to the right team. Let’s simplify and call that five steps.
At 0.005 AI Units per step, that is 0.025 AI Units per run.
- If it runs 100 times, that is 2.5 AI Units.
- If it runs 10,000 times, that is 250 AI Units.
And if the agent is not basic, or if the workflow has more steps, or if it runs more often than expected, the number changes quickly.
This is why AI Unit control is not only about protecting the company’s finances.
With usage-based AI, process inefficiencies start showing up in the bill. If an agent is built on top of a messy workflow, takes unnecessary steps, checks too many systems, or depends on poor data, the company may end up paying more without getting proportional value back.
That is why cost control and process design need to be part of the same conversation.
How to Design an AI FinOps Strategy Inside SAP
Controlling SAP AI Unit consumption is not just about checking SAP for Me at the end of the month.
By then, you may already know the problem.
You just found out too late.
A better strategy starts before the consumption happens. You need to understand which capabilities consume AI Units, who can use them, where usage is happening, which agents are running, and what process problem they are supposed to solve.
FinOps here is a fundamental part of the strategy. Not because companies need more process, but because they need a clear way to avoid one common problem: letting AI run across teams without enough control, only to discover later that a few workflows, users, or agents consumed a large part of the available units.
Now, let’s look at some of the areas teams should review to build a healthier AI consumption model inside SAP.
Layer 1: Separate Included AI from Premium AI
The first step is understanding what type of AI capability you are dealing with.
Not every Joule or SAP Business AI feature has the same commercial impact. Some capabilities are included as part of the standard product experience. Others are Premium AI capabilities and may consume AI Units.
SAP describes this distinction as Base AI and Premium AI. Premium AI includes more advanced capabilities, including Joule Agents and other Business AI use cases that are monetized through AI Units. For agents, SAP Learning explains that consumption can be measured by steps, with Basic, Standard, and Advanced Agents consuming different amounts of AI Units per step in consumption-based scenarios.
How to do this in SAP
Start by checking which Business AI capabilities are being activated and whether they are included or premium. The activation flow for Business AI packages is managed through SAP for Me, under Portfolio & Provisioning → Business AI, where packages can be activated and assigned.
For specific products, also review the product-level documentation because some embedded AI capabilities may have their own activation steps and AI Unit requirements.
What to watch
Before scaling a use case, ask a very simple question: is this included AI, per-user premium access, request-based consumption, or agent step-based consumption?
That answer changes the cost-control strategy.
Layer 2: Control Access Before Usage Spreads
The second layer is access.
This is basic, but it matters. Many cost issues start because access is opened too broadly before anyone understands the usage pattern.
For some Premium AI packages, assigned users directly affect cost. Joule for Consultants is a good example: SAP Learning says it uses a per-user-per-month model, charging 35 AI Units per assigned user per month. Each assigned user gets included requests, and overages are charged if total usage exceeds the pooled allowance.
How to do this in SAP
Use SAP for Me to activate the Business AI package, then assign users through the relevant identity and access management setup. SAP’s activation documentation points to SAP for Me for Business AI package activation, and user assignment depends on the specific package or product.
For Joule for Consultants, follow the specific provisioning and assignment documentation for that package.
A practical way to manage it
Start with the people who actually need the capability for a defined use case or project. Review assigned users monthly. Remove users who no longer need access. Keep test users and production users clear. Avoid assigning access “just in case.”
If the pricing model is tied to assigned users, unused access can still become part of the cost picture.
Layer 3: Create an Internal Allocation Model
SAP AI Units are generally consumed from a central pool. That means companies should not assume they can create perfect native “wallets” for each team and automatically block every team when it reaches its internal limit.
You may decide that part of the available AI Units should support consulting work, another part should support finance automation, another part should support procurement agents, and another part should support development use cases.
The important thing is to define that logic before consumption starts spreading.
How to do this in SAP
SAP for Me can show AI Units available, consumed, expiring, and consumed by product and feature through the Business AI tab.
For BTP-related services around AI workflows, SAP BTP cockpit provides Costs and Usage monitoring at global account, directory, and subaccount levels. SAP says you can compare usage and costs across services and subaccounts, view monthly trends, and drill down into subaccounts and service plans.
So the internal allocation model usually works by mapping your team or use-case budgets to the SAP products, AI features, BTP subaccounts, service plans, or environments where consumption is expected.
A practical way to manage it
Define your internal allocation by team, project, or use case. Map each allocation to the SAP product, AI feature, BTP service, or subaccount where usage will happen. Assign an owner. Monitor actual consumption in SAP for Me and BTP cost tools. Review the allocation each month.
The goal is to avoid one team, one experiment, or one badly scoped agent absorbing units that could have been used on stronger opportunities.
Layer 4: Use Budgets and Alerts, but Know Their Limits
Caps are important, but we need to be precise.
In SAP BTP, you can create budgets and alerts for usage or cost thresholds. SAP Alert Notification can trigger an event when a specified threshold for an active cost or usage budget in an SAP BTP global account has been exceeded.
That helps.
But it should not be confused with a perfect hard stop for every AI Unit scenario.
Alerts are usually a review trigger. They tell you something crossed a threshold. They do not automatically redesign the agent, stop the business process, or prove whether the consumption was worth it.
How to do this in SAP
In SAP BTP cockpit, use budget configuration for your global account where available. Enable thresholds and notifications so the right owner gets alerted when usage or cost exceeds the expected level. For subaccount-level tracking, open the relevant subaccount and use Usage Analytics to monitor actual usage data for services and applications in that subaccount.
How to think about caps
A cap does not need to block innovation.
It can create the right kind of pressure.
If AI Units are limited, teams have to prioritize better, avoid low-value experiments, and think twice before running agents on workflows that are not ready.
If a team hits its limit, the conversation should not be “stop using AI.”
It should be: “show what improved, and then we decide if this deserves more budget.”
Layer 5: Monitor AI Units in SAP for Me
SAP for Me is one of the main places to understand AI Unit consumption.
The Business AI tab gives customers visibility into AI Units available, units expiring, and consumption. SAP also notes that the current-month consumption can be an estimate until balances are finalized.
This is useful because it gives finance, IT, and SAP teams a common place to see the AI Unit pool.
But it is still visibility.
Not the whole control system.
How to do this in SAP
Go to SAP for Me → Finance & Legal → Business AI tab. From there, review AI Units available, consumption, expiration, and product or feature-level consumption where available. SAP also references AI Units in the Finance & Legal overview area.
What to watch
SAP has also stated in a KBA that AI Units are required for Premium AI usage in both test and production environments, and that SAP for Me currently aggregates AI Unit consumption per Premium AI feature without separating by environment type.
That matters.
If your team is testing Premium AI features, that testing may still affect consumption. So test activity should be planned and monitored, not treated as invisible.
Layer 6: Monitor Joule Usage Separately
SAP for Me helps with AI Units, but it may not show every detail of Joule usage.
SAP states in a public KBA that specific Joule usage data is not displayed in SAP for Me at this time. Instead, Joule Analytics Center provides visibility into Joule usage, including message and conversation usage in the production landscape.
SAP Help describes Joule Analytics Center as a tenant-specific dashboard that displays Product Usage, Scenario Usage, Interaction Type Usage, and Client Type Usage.
How to do this in SAP
Use Joule Analytics Center to review Joule usage patterns. Look at product usage, scenario usage, interaction type usage, and client type usage.
This helps answer a different question than SAP for Me.
SAP for Me tells you more about the AI Unit pool.
Joule Analytics Center helps you understand how people are using Joule.
What to watch
If Joule usage rises, that may be good. Maybe users are getting value.
But if usage rises in scenarios that do not connect to process improvement, you may need to rethink the use case, the training, or the way Joule is being positioned internally.
Layer 7: Design the Process Before the Agent
This may be the most important layer.
As we saw earlier, agent cost is not only about how many times an agent runs during a given period. It also depends on the number of steps the agent needs to complete a task, the systems it has to check, and the complexity of the workflow behind it.
How to do this in SAP
When designing or reviewing agents, look at the agent type, the number of steps, the workflow logic, the systems it needs to check, and how often it runs. For customer-built agents in Joule Studio or scenarios not included in a package, SAP Learning describes consumption by agent step and agent category.
What to watch
Before automating a process with an agent, review the process itself.
If the workflow has too many exceptions, messy data, unclear ownership, or manual approvals outside SAP, the agent may not fix the inefficiency. It may simply make the inefficiency more expensive.
For example, if an agent has to check five systems because master data is fragmented, that extra work may show up as more steps and more consumption. If a simpler workflow or better integration can solve the same problem, AI may not be the first thing to add.
Layer 8: Monitor the BTP Services Around the AI Workflow
Joule and SAP Business AI often sit next to other SAP services.
A real AI workflow may involve BTP, Integration Suite, SAP Build, SAP AI Core, APIs, subaccounts, and service plans. So even if the article is about Joule and AI Units, the surrounding BTP consumption also matters.
SAP BTP cockpit supports cost and usage monitoring at global account level, and lets users drill into subaccounts and service plans for detail. SAP also supports directory and subaccount usage analytics.
How to do this in SAP
Use SAP BTP cockpit → Costs and Usage to monitor current and historical usage. Drill down into directories, subaccounts, and service plans. Use labels where appropriate to filter by custom criteria.
For subaccount-level analysis, open the subaccount and go to Usage Analytics.
What to watch
If an AI workflow depends on integrations, APIs, automation, or other BTP services, do not look only at AI Units. Look at the full cost chain around the use case.
Sometimes the expensive part is not only the AI call.
It is the architecture around it.
Layer 9: Decide When Not to Use AI
This one is less about configuration and more about judgment.
Not every workflow needs Joule.
Not every automation needs an agent.
Sometimes a rule, a script, a workflow, a better integration, or cleaner master data will solve the problem more cheaply and with less risk.
How to do this in SAP
Before activating or building an AI use case, compare it against standard SAP capabilities, SAP Build automation, Integration Suite, workflow configuration, rules, and existing process options.
If the process is stable and rules-based, a traditional automation may be enough.
If the process requires language understanding, business context, exception handling, or decision support, Joule or an agent may be more appropriate.
What to watch
AI is powerful, but it should not become the default answer.
If a process is inefficient, putting an agent on top of it may expose that inefficiency in the bill.
Layer 10: Review Monthly and Reallocate
The last layer is the operating rhythm.
You need a monthly review where IT, finance, SAP owners, and business owners look at consumption together.
Not to punish usage.
To understand it.
How to do this in SAP
Use SAP for Me for AI Unit consumption, Joule Analytics Center for Joule usage, and SAP BTP Costs and Usage for services, subaccounts, and service plans around the AI workflow. Combine that with your internal allocation model.
What to review
- Which features consumed the most AI Units?
- Which teams or use cases were expected to consume them?
- Which agents or workflows grew faster than expected?
- Which alerts were triggered?
- Which use cases deserve more budget?
- Which ones need redesign?
- Which ones should stop?
The goal is to make sure AI Units do not disappear into black holes of consumption.
They should go to the use cases where the business can explain why the spend makes sense.
Where a SAP Partner Fits In
SAP provides the platform, the AI Unit model, and tools for visibility and estimation.
But as we saw during this article, controlling spend usually requires coordinating a lot of elements:
- Use-case selection,
- Agent and process design,
- Landscape integration,
- Dashboarding and monitoring,
- Governance,
- And AI FinOps.
That is where a SAP partner can help.
A partner can help separate good AI opportunities from expensive distractions. The sweet spot is not “spend less on AI.” It is knowing where the spend makes sense, where experimentation is useful, and where consumption is starting to move faster than the value behind it.
If that balance is missing, the risk is not only a bigger bill. It is that leadership starts losing confidence in AI initiatives altogether. And when that happens, even projects with real medium or long-term potential can get paused too early.
At Inclusion Cloud, we have been SAP partners for more than 20 years. Our certified technical and functional consultants can help you get more value from your SAP investment while keeping the financial side under control.
If your team is planning Joule initiatives or trying to navigate the new challenges of SAP tokenomics, we can start with a discovery call to understand your priorities and see how we can help.