As Enterprise Architects guide their organizations through AI discussions, they may notice that the same core questions commonly surface. One effective way to navigate these conversations is to analyze AI through recurring architectural perspectives. These perspective lenses help EAs clarify where business value emerges, how information moves through the system, and which technical decisions shape a secure and scalable AI landscape.
There are four distinct views that EAs commonly use:
- Business and Use-Case View: AI adds value inside SAP processes and supports business outcomes
- Data and Integration View: SAP data sources, SAP Datasphere / SAP HANA Cloud, events, APIs, and non-SAP data come together
- AI/ML Ops View: Models are trained, deployed, executed, and monitored using SAP AI Core, SAP AI Launchpad, or partner models
- Security and Compliance View: Identity, access, privacy, guardrails, and auditing apply across the AI landscape
Analyzed through these lenses, commonly experienced questions tend to follow clear patterns, as do the approaches used to answer them. Recognizing these patterns helps Enterprise Architects guide discussions more effectively and align AI decisions with long-term architecture principles.
Common Questions with Winning Approaches
1. "Should we use embedded SAP Business AI in the apps, or build our own custom AI?"
To approach this topic, a reliable starting point is:
- Use embedded AI for standard SAP processes — it’s built-in, governed, and aligned to best practices.
- Consider custom AI when the organization needs meaningful differentiation or cross-application orchestration.
- Always check whether extending standard AI use cases is possible before embarking on custom development.
2. "Where does generative AI fit into our landscape?"
GenAI excels in tasks such as knowledge retrieval, summarization, guidance, and assistance. Use retrieval augmentation to inject SAP-specific context so outputs remain accurate and business aware.
Platform and Architecture Choices
1. "Where should AI models actually run?"
- For most side-by-side AI extensions, SAP BTP is the recommended home.
- Use SAP HANA capabilities and ISLM only when the scenario requires on-stack deployment.
2. "How do we isolate tenants and environments?"
Apply SAP BTP subaccounts, role collections, and SAP AI Core resource groups to ensure proper separation and governance.
Data Architecture and Integration
1. "How do we safely access SAP application data for AI?"
- Always use SAP APIs (OData, CDS views).
- Avoid direct database access under the application layer.
- For side-by-side scenarios, leverage SAP HANA Cloud and SAP Business Data Cloud.
2. "Which data layer should we use for AI — SAP Business Data Cloud, SAP HANA Cloud, or both?"
- It depends on the user, the use case, and the technical requirements.
- The right data layer is the one that best supports the AI pattern you’re implementing.
Security, Privacy, and Compliance
"How do we manage identities end-to-end when AI is involved?"
Follow SAP’s recommended identity architecture for cloud services and SAP IAM Integration with SAP Cloud Identity from the SAP Architecture Center.
Even with a well-defined reference architecture in place, EAs often encounter the same recurring challenges across many organizations. Three of the most common challenges are the following:
- Building custom AI before checking embedded capabilities.
- A common risk is moving into custom development before assessing whether SAP applications already offer embedded AI that addresses the same process.
- Unclear data ownership or weak semantic definitions leading to poor model performance.
- When data isn’t well-governed or lacks a consistent semantic layer, even strong models struggle to produce reliable results.
- Calling external LLMs without guardrails, logging, or contractual protections.
- Direct, non-managed use of external foundation models or LLMs can expose organizations to compliance, privacy, and audit risks.
Recommended Best Practices
To avoid pitfalls, EAs typically rely on proven architectural approaches.
- Embedded-first best practice: Begin with SAP-delivered AI features inside the applications. Measure impact and only extend or customize when a business requirement exceeds standard capabilities.
- Side-by-side extension best practice: Build AI applications on SAP BTP that:
- Consume SAP APIs or events
- Manage governed context in SAP HANA Cloud or SAP Business Data Cloud
- Run models through SAP AI Core
- This approach keeps extensions clean, governable, and closely integrated with the surrounding business processes.
Example Stakeholder Questions and Answers
When discussions transition from theory and concepts to implementation, specific practical questions may arise. Enterprise Architects typically address implementation questions with the following responses:
1. "Can we bring our own model?"
Yes. Organizations can bring their own models through SAP AI Core, where they can register, deploy, and operate them with governance support. The key is to verify that the model's licensing terms and frameworks are supported.
2. "What is the impact on SAP S/4HANA?"
To keep the digital core clean, EAs recommend using side-by-side extensions that interact with SAP S/4HANA through standard APIs. This prevents the introduction of custom AI code into the core system.
3. "How do we manage upgrades?"
Models should be treated as versioned assets with proper change control. Align model lifecycle management with SAP’s application release cycles to avoid disruptions and ensure consistent behavior across environments.
Lesson Summary
AI architecture discussions within organizations often follow predictable patterns. By viewing SAP Business AI through business, data, platform, and security perspectives, you can quickly recognize recurring questions and address them with proven SAP-aligned approaches. You are now equipped to guide AI architecture decisions within your organization through architectural decisions, avoid common pitfalls, and steer conversations toward secure, scalable, and value-driven AI implementations.