AI is not an end in itself - which is why SAP partner Snap organizes so-called experience workshops on numerous SAP topics. In 2026, there will be two experience workshops in collaboration with E3 magazine for the AI challenge: on April 23 in Heidelberg at the Steampunk and BTP Summit and on June 11 in Salzburg at the Competence Center Summit with a focus on CCoE.
E3: Which topics should existing SAP customers currently consider from the user's perspective?
Christian Knell, Snap Consulting: For users, the focus is primarily on three AI topics: Joule for Business, the use of Joule Agents directly in business processes and the basic and premium AI features provided by SAP. The aim is to automate routine tasks, provide better support for decisions and provide contextual support for process steps.
E3: Fo whom is Joule for Consultants suitable?
Knell: SAP is now positioning Joule for Consultants as the preferred tool for key users, consultants and in-house consultants. The added value: the tool bundles knowledge from the SAP knowledge base, including SAP Notes, SAP documentation, blogs and other official sources, and simplifies access to it. It is also generally advisable to use a general-purpose AI tool in the company, such as ChatGPT, Claude or Microsoft 365 Copilot. By integrating relevant data sources with the help of MCP servers,
For example, SAP documentation or Atlassian Confluence, these systems are given the necessary context to provide relevant and correct answers.
E3: What issues are there from the perspective of the IT development department?
Knell: The development of your own Joule skills and agents with Joule Studio. This allows customer-specific SAP applications and third-party systems to be integrated into the Joule ecosystem. The extension of the agents delivered by SAP is also relevant. Existing and new applications can be extended via AI functionalities using the Abap AI SDK or LangChain (UI5, CAP).
E3: And SAP BDC?
Knell: The connection of relevant data sources via SAP Business Data Cloud, BDC, is interesting. This ensures that Joule has access to the relevant company-specific context beyond S/4 - SAP Datasphere, SAP Analytics Cloud, SAP Business Warehouse, etc.
E3: How can an existing SAP customer approach the topic of SAP Business AI?
“The best way to get started with SAP AI is through a combination of practice, self-learning and targeted training.”
Christian Knell,
CEO,
Snap Consulting
Knell: The best way to get started with SAP AI is through a combination of practice, self-learning and targeted training. The SAP tutorials in the Learning Hub are a good first step. To try out the learning content in practice, SAP provides the necessary infrastructure as a free-tier offering.
E3: Then there is the Discovery Center: Learning Journeys?
Knell: Yes, the SAP Discovery Center offers learning journeys that use a project board to guide you through topics, use cases and products, conveying content and creating practical references.
E3: In addition, the Snap experience workshops, each with a focus, will be offered at the SAP Community Summits in Heidelberg on April 23 and in Salzburg on June 11 of this year.
Knell: The hands-on workshop is a quick, practical introduction to the basics, concrete use cases and real project experience.
E3Are there any hidden costs and unforeseen fees such as AI units, consumption, etc.?
Knell: The costs incurred are always presented transparently by SAP - for example in the Discovery Center or in SAP Notes explicitly provided for this purpose (3437766, Availability of Generative AI Models - SAP for Me). The use of virtual currencies such as AI units, which have to be converted into euros using a conversion factor, takes some getting used to. This depends on the conditions that a customer has agreed with SAP for the purchase of the units. SAP also offers an AI feature estimator in the Discovery Center for the consumption of AI units by services. Consumption-based billing must be checked regularly, see Governance, especially intensively in the first weeks and months of use. Regularly thereafter nonetheless. You should also regularly check whether services that you have subscribed to are actually being used. There can be one or two surprises with a billing document from SAP.
E3: How does SAP ensure data sovereignty and GDPR compliance when sensitive ERP data is transferred to external hyperscaler LLMs?
Knell: SAP emphasizes „Responsible AI“ with a focus on ethics, security and compliance and refers to corresponding AI governance for solutions such as Joule, AI Core and AI Launchpad. Furthermore, it is assured that customer data will not be passed on to third-party LLMs for model training use or stored permanently. Security is ensured via the SAP contractual framework and DPA, including technical and organizational measures such as TOMs and GDPR mechanisms such as SCC/IDTA and subprocessor transparency, which can be found in the SAP Trust Center. Additional data protection measures such as data masking, content filtering and masking as well as pseudonymization are available for AI Core and GenAI Hub.
E3: Is SAP BTP mature enough as a foundation for AI innovation to be a long-term enterprise platform?
Knell: For many companies, SAP software is already the central platform for business-critical applications. This is a strong argument in favor of SAP BTP, namely to provide uniform access to AI. In particular, integration with other AI platforms is challenging in practice, as data, identities, authorizations and processes must interact seamlessly across system boundaries. Another added value is the coverage of typical company requirements such as security, compliance, organizational and regulatory guard rails that need to be taken into account. This reduces the workload for customers because integration, governance and operation do not have to be implemented anew for each model provider.
E3: Doesn't the generative AI hub lead to an additional vendor lock-in, as the existing customer is tied to SAP for the use of third-party LLMs?
Knell: The GenAI Hub serves as a central service in the enterprise architecture and thus makes it easier to change or upgrade the AI models used in custom code, as it masks API specifics of LLMs. This also creates an architectural dependency. However, if there is an urgent need to connect an LLM or AI tools from other providers in customer-specific coding, this is also possible without using AI core via API consumption, such as web services. However, this also makes the IT landscape more complicated.
E3: Why are key AI innovations only available for the SAP cloud?
Knell: SAP communicates this openly. The strategic focus is clearly on the cloud business. As a result, the vast majority of innovations are only made available in the cloud. However, some features can also be used by on-prem customers, such as the connection of AI models for customers' own developments via the Abap AI SDK and Generative AI Hub on the BTP.
E3: What about reliability and traceability, i.e. explainable AI?
Knell: Until now, SAP has used a „human in the loop“ approach for activities that involve changes to the system by Joule Skills or agents. If an agent comes to the conclusion that bookings need to be made, explicit confirmation by the user is required. For AI functionalities delivered by SAP that are based on the Generative AI Hub or AI Core, techniques are used to ground the LLMs, for example Retrieval Augmented Generation. Additional context is used to counteract hallucinations. The customer is responsible for in-house developments. In addition to using Hana's Vector DB functionality to provide additional context, the LLM can also be provided with calls to tools, such as methods implemented in Abap, for critical calculations or investigations. The traceability and explainability of agents should also be ensured via monitoring functions. Authorizations are propagated from the called SAP systems to the SAP identity provider in such a way that no user can carry out activities via Joule that are also prohibited manually due to missing authorizations.
E3: Does SAP prevent „astronomical“ damage from Agentic-AI and Joule Agents?
Knell: The „Human in the Loop“ approach shifts responsibility to the users, which is at least one hurdle to overcome. Otherwise, as with every software component delivered by SAP: Bugs are fundamentally possible and can have an impact on productive data and processes. These processes must be taken into account in the course of upgrade projects.
E3: Are there realistic GenAI use cases in S/4 for existing customers?
Knell: The range of functionalities provided can be viewed in the SAP Discovery Center, where you will find specific approaches - as with any investment, it is important to check in advance whether this can offer an ROI in the specific customer situation. We have had very positive experiences with customers using Document AI, the Recommendation Service or AI-based machine translations via SAP Translation Hub. Another point is the use of agents, which will automate many processes in the future. We are noticing the demand for automation using agents in greenfield projects.
E3: How realistic are citizen developers without abap code for business-critical AI apps such as
SAP Build Code and Joule Studio?
Knell: The same applies as for low-code/no-code developments of conventional applications. Compared to pro code, the scope of possibilities is certainly limited. In many cases, despite the use of low-code/no-code tools such as Joule Studio, in-depth technical knowledge is an advantage. The hurdle for citizen developers is often not the tool, but the structural approach and logic for the development of program sequences. However, SAP is also planning pro-code agents on the BTP, which can be used to cover requirements that go beyond the possibilities of the low-code/no-code approach.
E3: And Abap's own developments?
Knell: The Abap AI SDK is ideal for in-house developments. This is available from S/4 with SAP_BASIS 7.56. This allows models provided in the Generative AI Hub to be called up directly from Abap. The Abap AI SDK also offers functionalities for calling Abap methods by the AI. This is also possible in the UI5 and CAP environment, for example using libraries such as LangChain. The extent of the necessary interventions in the code depends heavily on the structure and quality of the existing programs. Alternatively, any AI tools or LLMs can be integrated into existing Abap applications via a web service.
Continue to the partner entry:
DE 
EN 
ES 
FR 
