SAP's Prebuilt AI Services and Machine Translation on SAP BTP: Status, Successors and Terminology Control

What you will learn

• Read the status of Data Attribute Recommendation, Personalized Recommendation and Translation Hub from the dated SAP guides and state the run-off rule
• Call the Document Translation API synchronously and asynchronously and choose between the default engine and `model=llm`
• Protect terms with `translate="no"` and a term-aware glossary and verify the result on the output text
• Place the orchestration translation module in a generative AI hub pipeline and name its order relative to masking
• Decide between Document Translation, a generative AI hub model, SAP-RPT-1 and Document AI for a given request

Concept overview

Answer first. As read on 5 November 2026, the classic row of "SAP AI services" on SAP BTP has split in two. Data Attribute Recommendation and Personalized Recommendation are in run-off: both guides carry a caution that the service was removed from the list of Eligible Cloud Services on 31 March 2026. SAP Translation Hub, with its Document Translation service, is still documented as a working service and has gained an LLM option next to SAP's own machine translation. Document AI, the fourth service on the old slide, belongs to M328 and is not repeated here.

Prerequisites

• An SAP BTP account with a Document Translation service instance and service key (a free or trial plan is enough for the lab)
• Basic knowledge of REST calls and OAuth bearer tokens; curl is enough
• For the optional orchestration step: M325 (generative AI hub) and an SAP AI Core instance on the extended plan

Outcomes

• Quote the removal caution of a prebuilt service correctly: date, what continues, what stops and what the guide does not say
• Explain why SAP-RPT-1 can replace Data Attribute Recommendation use cases but is only a partial answer for Personalized Recommendation
• Reproduce the lab: four translation calls and one invalid glossary call, with protected term, glossary effect and error code documented
• Name the two ordering rules of the orchestration translation module and their privacy consequence
• Write acceptance tests for translation that assert on output text rather than on HTTP status