AI needs context and controls
AI can produce impressive results, but it does not automatically understand a company's business processes, policies, data, or system configuration.
A general-purpose AI model may understand the concept of a purchase order, but it may not know:
- The customer's approval thresholds
- Which fields are mandatory
- How organizational structures are configured
- Which process variations are in scope
- Whether a proposed solution follows company policy
- Whether the information it generated is factually correct
- Which actions it is authorized to perform
The quality of the output depends heavily on:
- The quality of the training data
- The information provided to the model
- The clarity of the prompt
- The system's access to relevant business context
- The tools and permissions available to it
- Human review and validation
These considerations become even more important with agentic AI because the system may be taking actions, not merely generating suggestions. Organizations therefore need clear controls around:
- Data access
- User authorizations
- Audit trails
- Approval requirements
- Segregation of duties
- Exception handling
- Human oversight
- Accountability for decisions
This is why implementation expertise remains essential.
AI can accelerate research, documentation, analysis, content creation, and process execution, but consultants still need to validate whether the output and actions are accurate, relevant, compliant, secure, and appropriate for the customer's solution.
What makes SAP's AI different than standalone third-party AI tools?
SAP's AI capabilities differ from a standalone third-party AI tools because they operate within the context of SAP applications, business processes, and data.
SAP has spent more than 50 years developing software for a wide range of industries and business functions. That experience is reflected in the process logic, data structures, and business terminology built into SAP systems. Today, SAP applications support more than 120 mission-critical business processes.
This provides SAP's AI capabilities with three important forms of context:
- Process and industry content
- Semantically-rich business data
- Enterprise governance
Process and industry content
The models and services draw on the processes and industry-specific concepts already represented in SAP applications. This helps the AI interpret a request in relation to the business activity being performed, rather than treating it as a general language question.
For example, terms such as business partner, purchase order, goods receipt, and company code have specific meanings and relationships within an SAP system.
Semantically-rich business data
Business data is more useful to AI when the system understands what the data represents and how different data objects relate to one another.
The SAP Knowledge Graph organizes approximately 7.3 million data fields and connects them with business meaning. Rather than working only with raw values, AI can use information about the underlying objects, relationships, and process context.
For example, the system does not only see an identification number. It can recognize that the number represents a supplier, that the supplier is associated with a purchase order, and that the purchase order is connected to a goods receipt and an invoice.
Enterprise governance
AI used in business processes must operate within the same types of security, quality, compliance, and access controls as the rest of the enterprise system.
SAP applies governance controls to its AI workloads and maintains certifications related to AI management, information security, and quality management, including ISO 42001, ISO 27001, and ISO 9001.
These controls are part of the platform and help determine which data the AI can access, which actions it can perform, and where human review or approval is required.
The main distinction is therefore not simply that SAP provides an AI model. It is that the AI can operate with the process definitions, business data, relationships, authorizations, and governance already present in SAP applications.
AI makes it possible to use this existing business context in new ways. For example, to explain information, generate content, identify patterns, recommend actions, or support the execution of a business process.
Practical Example: Blocked Supplier Invoice
Consider a user who asks:
Why is this supplier invoice blocked, and what should I do next?
A general-purpose AI tool understands the language of the question and may understand common accounts-payable concepts.
It could provide a response such as:
The invoice may be blocked because of a price variance, quantity variance, missing approval, duplicate invoice, or mismatch with the purchase order. Review the invoice, purchase order, and goods receipt to identify the cause.
This may be useful as general guidance. However, the AI does not automatically know:
- Which invoice the user is referring to
- How the customer's SAP system is configured
- The actual blocking reason
- Whether a purchase order or goods receipt exists
- The customer's tolerance limits
- Which user is responsible for resolving the issue
- Which actions the current user is authorized to perform
To answer more specifically, the user would need to copy information from the ERP system into the third-party tool. That may also create security, confidentiality, or data-governance concerns, depending on the information involved and the tool being used.
SAP AI can operate within the context of the SAP application and the business process the user is currently performing.
Depending on the specific capability and the user's authorizations, it may have access to relevant business objects such as:
- The supplier invoice
- The purchase order
- The goods receipt
- Supplier master data
- Approval status
- Payment status
- Configured tolerance limits
- Related workflow tasks
It could therefore provide a more contextual response, such as:
Invoice 5100048721 is blocked because the invoiced quantity is 100 units, while the posted goods receipt is for 80 units. The difference exceeds the configured quantity tolerance. Twenty additional units must be received, or the invoice quantity must be corrected before the invoice can be released.
The system could also direct the user to the relevant document or application, recommend the appropriate next step, or initiate an authorized workflow.
The difference is not necessarily that the underlying language model is better at explaining invoice processing. The difference is that SAP AI can combine language capabilities with the business context already available in the SAP system.
Note