Amazon QuickSight's Multi-Dataset Topics Let Analytics Teams Query Across Data Without Rebuilding Pipelines

Amazon QuickSight now lets teams merge multiple datasets into a single conversational Topic, removing one of the more persistent bottlenecks in self-service business intelligence: the need to pre-join or flatten data before anyone can ask a question of it.

The feature, called Multi-Dataset Topics, is detailed in a new AWS Machine Learning blog post that lays out best practices for combining data sources inside QuickSight's natural-language Q&A interface. Rather than forcing analysts to build a single denormalized table upfront, the capability gives teams two distinct paths for stitching datasets together, each suited to a different kind of analytical need.

Two Ways to Combine Data

The first approach relies on defined relationships. Here, QuickSight builds a logical join graph from explicit relationship keys set between datasets, and at query time it executes inner joins across them. This path supports up to 12 datasets in a single Topic, making it a natural fit for well-structured, relational data where the join logic is stable and predictable.

The second approach hands more of the reasoning to AI. Instead of fixed join paths, it uses Chat-powered SQL generation informed by a semantic layer, letting the system interpret a user's question and construct the appropriate SQL on the fly. According to AWS, this method isn't limited to simple inner joins: it can produce outer joins, unions, subqueries, self-joins, cross-grain comparisons, and conditional join logic, all without being locked into a fixed structural template. That flexibility matters for scenarios where the relationship between datasets shifts depending on the question being asked, something a static join graph can't easily accommodate.

Semantic Layers Do the Heavy Lifting

What makes the AI-generated SQL path reliable, rather than a black box guessing at table relationships, is the semantic layer QuickSight builds around each Topic. This includes dataset-level custom instructions, Topic-level instructions, field synonyms, and field descriptions, all of which give the underlying model context about what the data actually means and how it should be interpreted. Rather than inferring intent purely from column names or raw schema, the system is guided by human-authored metadata that clarifies ambiguous terms and business logic before a query is even generated.

Why It Matters

For analytics teams, the practical upside is a shorter path from raw data to answers. Multi-Dataset Topics reduce the engineering overhead typically required to prep data for self-service dashboards, since teams no longer need to physically merge sources into one dataset before end users can query across them conversationally. The tradeoff between the two methods is really a tradeoff between control and flexibility: relationship-based joins offer predictable, auditable query behavior at a defined dataset ceiling, while AI-generated SQL trades some of that predictability for the ability to answer a far wider range of ad hoc, cross-dataset questions.

AWS positions the feature as part of QuickSight's broader push toward natural-language analytics, where the semantic layer, rather than manual query writing, becomes the primary interface between business users and complex, multi-source data.