Prompt Engineering for SAP Consultants: Better Results from Joule and General-Purpose LLMs

Keep this: job aid

SAP prompt card

• Six parts: role, task, context (release, process, data), constraints, 1 to 5 examples, exact output format.
• Mark data with tags or headers; long documents first, the question last.
• Give reasons for rules ("because testers will execute this in a regulated system").
• In Joule: name the object and the outcome, @mention the assistant, and read the Source and AI Reasoning section.
• Tell the model to mark missing SAP facts as [CONFIRM] instead of inventing T-codes, apps or message numbers.
• Ask for quotes and citations; allow "I don't know"; restrict to supplied material.
• Verify SAP facts in SAP Help, the Simplification List, the Fiori apps reference library and the system; recompute numbers.
• Never paste credentials, bank details, national IDs, HR or health data. Mask names, contacts and user IDs.
• Use approved channels: Joule in SAP apps, the enterprise LLM, or a governed generative AI hub app with masking.
• Save working prompts as versioned templates with test inputs; re-run them when the model changes.