Making NLQ More Explainable, Reliable, and Actionable

[ThierryAudas]

In a previous Data Professionals discussion, “NLQ in the real world: what metadata setup made it successful for you?”, I asked the community what really makes natural language query work in practice, from business-friendly labels and multilingual metadata to clearer model semantics and less ambiguous terminology. The discussion reinforced a practical point: successful conversational analytics depends not only on AI models, but also on the quality, consistency, and readiness of the underlying business metadata.

This post follows up on that discussion. With the Q3 2026 release of SAP Analytics Cloud, several enhancements address the practical issues practitioners raised by making it easier to understand how questions are interpreted and where a model may limit conversational analytics accuracy. Kudos to @Flavia_Moser for driving these improvements.

Explain How Questions Translate into Analytics

One of the most common requests from customers has been greater transparency into how natural language questions are interpreted. With the latest enhancements to the Just Ask feature, Analytics Cloud now provides AI-generated explanations that help users understand how a request was translated into an analytical result.

When a user asks a question such as “Show me sales by brand for this year,” Analytics Cloud now explains how the request was interpreted, shows which filters, variables, and visualization choices were applied, and generates a headline that more accurately describes the result. It also provides a concise business summary of the generated insight, helping users quickly understand not just the answer, but how the answer was produced.

This addresses a common trust gap: when users can see which fields, filters, variables, and chart choices were used, they are better equipped to validate the answer, spot misinterpretations, and refine the question when needed.

 NLQ explanation with the Just Ask feature

Check Model Readiness

Great analytics starts with a well-designed data model. To help users get more value from the Just Ask feature and Analytical Insights in Joule, Analytics Cloud now introduces model readiness checks for NLQ. This AI-assisted capability automatically evaluates data models and identifies common issues that can affect natural language analytics, including missing semantics, unclear hierarchy definitions, ambiguous labels, and other modeling gaps that make questions harder to interpret consistently.

The readiness assessment highlights potential issues and their severity, recommends corrective actions, and provides guidance directly within the modeling experience. In practice, this helps analytics teams prepare their data models for better AI outcomes before business users start relying on conversational analytics.

For organizations expanding self-service analytics, model readiness checks give modelers a more actionable way to improve NLQ quality before rollout, especially in areas where missing semantics, unclear hierarchies, or ambiguous labels can lead to inconsistent answers.

 Model readiness report in Analytics Cloud

 Expand Semantic Understanding

This release also broadens semantic understanding across data model types, allowing Analytics Cloud to better interpret user intent and generate more relevant insights.

These enhancements include fiscal calendar support with week-level granularity for models acquired in Analytics Cloud, compound dimension support for SAP Datasphere models, and external live version support for seamless planning models. In practical terms, this helps NLQ better respect the way business data is actually modeled, such as fiscal reporting periods, compound keys, and planning versions.

Make Analytical Insights in Joule More Actionable

Finally, this release also advances conversational analytics within Joule. Users accessing analytical insights through Joule in SAP applications such as SAP S/4HANA and SAP SuccessFactors can move from a natural language question to an interactive chart, a table view, or a recommended follow-up question without leaving the flow of work.

This changes the workflow from a static answer to a guided analysis path: business users can sort, rank, inspect details, and continue exploring related questions directly in the conversation experience.

Analytics Through Conversation

SAP’s vision is straightforward: enable business users to access and interact with timely insights through natural language, wherever they work. The enhancements in this release bring that vision closer to practice by making AI-generated analytics more transparent, helping teams validate and improve model readiness, and making insights easier to access through Joule in the context where decisions are made.

For practitioners, the important shift is not just that conversational analytics can answer more questions, but that the system now provides more context for validating those answers and more guidance for preparing the models behind them. As you evaluate these capabilities, it will be useful to test them against the same kinds of metadata, label, hierarchy, and terminology challenges discussed in the earlier Data Professionals thread. I’d be interested to hear which enhancements make the biggest difference in your own NLQ scenarios, and where further improvements would help.