Natural-Language Query over SAP Data — Text-to-SQL Patterns and Limits

What you will learn

• Route three real client questions to Just Ask, Joule SQL generation in HANA Cloud or a Genie Agent, and state whether the asker ever sees SQL
• Create a Genie Agent with the click path Genie Agents → New → Configure → Data / Settings → Share, within its 50-object and 100-instruction limits
• Declare a one-to-many join and a trusted SQL expression so that a header-value question returns 2,300 EUR instead of 5,100 EUR
• Write the six-point tool definition for an agent-callable query tool: audience, read-only role, pinned schema, SQL returned, audit log, benchmark attached
• Build a 20-question benchmark with ground truth and run it as a release gate after every configuration change
• Separate the TechEd 2026 items that are phased GA from those planned for Q1 2027 before promising them in a design document

Who this is for. A client has seen a demo where someone types a question and a chart appears, and now expects you to explain precisely what happened between the two. This module compares the natural-language-to-query systems you will meet in an SAP landscape — SAP Analytics Cloud's Just Ask, Joule's SQL generation in SAP HANA Cloud, and Databricks Genie inside SAP Business Data Cloud — and gives you a six-step Genie click path, a 20-question benchmark method and a six-point tool-definition checklist you can use on Monday. It builds on “AI-Ready Data — Quality, Semantics and Lineage” and “The Semantic Layer as LLM Context”; Genie facts are taken from the Databricks Genie documentation read on 28 Oct 2026.

1. Two audiences, two products, one pattern

Text-to-SQL is a pattern: natural language in, a structured query out, executed against governed data. SAP and its partners ship it three times (Just Ask, Joule SQL in HANA Cloud, Genie in BDC), for two audiences (see SAP Architecture Center — Analytical Insights in Joule for the Just Ask pipeline). Pick the system from the audience, not from the demo.

Prerequisites

• The Semantic Layer as LLM Context or working knowledge of analytic models and business-layer design
• AI-Ready Data or understanding of catalog metadata, glossary terms and authorisation scoping
• Familiarity with SAP Analytics Cloud and, ideally, exposure to a Databricks or SAP Business Data Cloud environment

Outcomes

• Correctly route a client's natural-language query need to Just Ask, Joule SQL generation or Databricks Genie based on audience and risk.
• Configure (or specify for a client's team) the metadata layers a text-to-SQL system actually reads, not the prompt wording.
• Build and defend a benchmark question set as the basis for judging any configuration change.
• State and enforce the human-review rule for any generated SQL capable of writing data.