AWS Demonstrates Cross-Platform Fraud Detection With SageMaker Data Agent

AWS has published a Big Data Blog post showing how its Amazon SageMaker Data Agent can be used to investigate fraud across data stored in both Snowflake and AWS. The post, published on 22 June 2026, walks through an end-to-end fraud-investigation scenario inside SageMaker Unified Studio notebooks.

The demonstration highlights three capabilities working together: running SQL analytics on Snowflake data sources, managing materialized views, and creating interactive charts. Used in combination, they let analysts query Snowflake alongside AWS data, pre-compute and schedule repeated aggregations, and produce visualisations from natural-language prompts — all in a single notebook, without writing boilerplate code or switching between tools.

One nuance worth noting: the capabilities themselves became available earlier, on 3 April 2026. The June post is a use-case walkthrough rather than the original launch announcement, so it should not be read as the general-availability date.

The first capability lets the Data Agent connect to Snowflake data warehouses through connections registered in SageMaker Unified Studio. The agent discovers available Snowflake databases, browses schemas progressively, and generates Snowflake-dialect SQL — including Snowflake-specific syntax such as FLATTEN, VARIANT column access, and semi-structured data handling.

Previously, the Data Agent generated SQL for AWS-native engines including Amazon Athena, Amazon Redshift, Apache Spark, and DuckDB, but it could not yet produce Snowflake-dialect SQL.

AWS frames this against a familiar pain point: until now, analysts had to write Snowflake SQL manually and export the results as CSV files to join with AWS data — a process that, by AWS's account, consumed one to two hours before any actual investigation could begin.

The second capability is materialized view management. Analysts describe the aggregation they want in plain language, and the agent generates the Spark SQL DDL, including SCHEDULE REFRESH syntax.

Materialized views store pre-computed results in Apache Iceberg format for fast repeated access, turning expensive full-table scans into sub-second queries. AWS notes the problem this addresses for fraud teams, where investigations can start with a 30-minute wait for queries that ran identically the day before.

The agent can also be asked to review existing work. A prompt such as "analyze my notebook and suggest which queries would benefit from materialized views" prompts it to recommend optimisations, create the views, and set refresh schedules.

The third capability changes how results are visualised. Instead of generating matplotlib code that produces static images, the Data Agent now creates native interactive chart cells powered by Vega-Lite.

These support bar, line, scatter, pie, area, heatmap and more, rendering inline with hover tooltips, zoom, and filtering. A prompt like "plot monthly revenue trends by region for 2025" generates an interactive chart directly in the notebook.

According to AWS's official announcement, the features are available in all AWS Regions where Amazon SageMaker Unified Studio is supported.

The common thread in this announcement is the cost of fragmented data. AWS's own framing shows analysts losing one to two hours exporting and re-joining data from separate platforms, and waiting half an hour for queries they had already run the day before. The new capabilities are essentially about closing the gaps between where data lives and where decisions get made.

That lesson applies well beyond fraud teams and well beyond AWS-specific tooling. Whether you run a professional services firm, a retail business, or a healthcare practice, the same friction shows up whenever your customer records, financials, and operational data sit in disconnected systems that someone has to stitch together by hand.

This is the foundation Intellova focuses on: bringing business data from many sources into one unified database, so the slow, manual work of joining and exporting is done once rather than every time a question comes up. With a clean, connected data layer in place, the kind of fast querying, smart aggregation, and AI-assisted analysis showcased here becomes far easier to put to work — on whatever questions matter most to your business.