Evaluating an End-User Workflow Example

A request can start in different ways, depending on how the user or system interacts with SAP Business AI:

• Via Joule – the copilot across SAP solutions
• Via embedded AI features – for example, in SAP S/4HANA, SAP SuccessFactors, SAP Ariba, and many others

A financial end-user submits the following into Joule:

"Show me all overdue invoices for customer [ACME] and summarize any trends."

With this single sentence, the end-to-end AI flow begins.

Before any data is distributed, SAP’s identity services ensure that the appropriate authorized persons are requesting elements:

• The user is authenticated through SAP Identity Authentication Service (IAS)
• The user’s authorizations are verified through SAP Identity Provisioning Service (IPS)

If both authentication and authorization are successful, the request proceeds. This ensures that AI does not return data that the user is not authorized to view.

Next, the necessary business context is organized to provide an accurate answer to the requested question. Depending on the use case, the following may be employed:

• SAP HANA Cloud Vector Engine: used for semantic search and Retrieval-Augmented Generation (RAG) over embedded content
• SAP Knowledge Graph: to understand relationships between entities such as customers, invoices, materials, and organizational structures.
• For Enterprise Architects: This set of steps displays SAP’s strength of bringing into business context applications, data and AI. The request is not just using a model; it is enriched with domain-specific structure and semantics retrieved from SAP systems.

Joule’s core engine now interprets what the user wants:

• It understands the natural-language request
• It determines user intent (for example: "analyze overdue invoices for a specific customer and highlight trends")
• It decides which SAP systems, skills, or agents need to be involved

Based on this understanding, Joule routes the request to the appropriate targets — for example, finance applications, document stores, or specialized agents that work on receivables and trends.

Once intent and context are clear, the request enters SAP AI Foundation on SAP BTP.

In this layer:

• The system selects the appropriate model – an SAP model, a partner large language model (LLM), or an embedded AI capability
• Execution is handled securely via SAP Generative AI Hub
• Guardrails and orchestration rules are applied to control how the model is used and how data flows

The selected AI model executes the request:

• Large language models (LLMs) answer questions, summarize information, or generate narratives
• Custom models handle domain-specific or organization-specific logic
• SAP Business AI services execute pre-built use cases delivered with SAP applications

Throughout this step, no enterprise data is stored or used for model training. Data is processed to generate the response but not reused to train underlying models.

The system then assembles a business-ready response:

• Combines insights from all involved SAP systems and AI services
• Validates and formats the results so they are clear and consumable
• Enriches the output with business context such as roles, business objects, units, and currencies

The goal is not just to answer the question, but to present information in a way that fits how SAP users work.

From the first prompt to the final answer, security and governance are continuously applied:

• Data remains protected according to SAP’s security and privacy standards
• All actions are logged, providing traceability and auditability
• Governance policies and guardrails are enforced automatically, including for cross-system agent orchestration

This is especially important for regulated industries and for scenarios where multiple agents act across processes and systems.

Finally, the user sees the result in their working context:

• The answer is displayed in Joule as a conversational explanation
• It is enriched with navigation links into the relevant SAP applications
• It is adapted to the user’s role and current UI, so follow-up actions are only a click away

Joule returns:

• A summary of overdue invoices for the customer
• A trend analysis (for example, "overdue amounts have increased over the last three months")
• A direct link to the relevant SAP S/4HANA screen where the user can drill down into individual invoice details

For the end user, this feels like a simple, guided, outcome-oriented interaction. For Enterprise Architects, it’s a concrete illustration of how Joule, agents, AI Foundation, and the broader SAP Business AI Reference Architecture collaborate to deliver secure, contextual, and deeply integrated AI experiences.

Understanding this flow helps architects explain how SAP Business AI supports real business tasks, not just isolated AI features.

Discover more information with SAP Architecture Center - Joule in SAP S/4HANA Cloud Private Edition and SAP S/4HANA Cloud Public Edition.

A simple user prompt can trigger a sophisticated, governed AI flow across SAP systems. By walking through an end-to-end example, you have seen how Joule, agents, data services, AI Foundation, and security controls collaborate behind the scenes. You are now equipped to translate abstract architecture concepts into a concrete, outcome-oriented user journey that clearly demonstrates SAP Business AI in action.