AWS customers expect fast, informed support when they’re evaluating new technologies, troubleshooting production issues, or planning cloud transformations. To deliver that experience at scale, AWS technical field teams need immediate answers to complex operational questions: Where is customer demand increasing? Which teams have the right expertise to respond? Are customer engagements being resolved quickly enough? And where are emerging gaps that could impact customer outcomes?

The AWS Technical Field Communities (TFC) program supports hundreds of thousands of these customer engagements annually across dozens of specialized technology domains. For program leaders and field teams, understanding the pulse of these engagements isn’t just about tracking metrics; it’s about making sure that we have the right skills in the right places at the right time to help our customers succeed. Yet, as the scale of these engagements grew, so did the complexity of the questions our leaders needed to answer. Traditional, static dashboards began to struggle under the weight of sophisticated, multi-dimensional inquiries. Stakeholders found themselves navigating a maze of different systems, manually cross-referencing datasets just to get a clear picture of how to better serve the customer. Getting to the “why” behind the data isn’t always a hard technical problem, it’s a workflow problem. A leader’s question becomes an interruption for a BI engineer, who pauses planned work, runs the aggregation, and returns an answer that inevitably spawns the next question. The real time lost isn’t in the query. It’s in the handoff between the person with the question and the person with the tools to answer it. Leaders were asking complex, real-time questions that crossed organizational and technical boundaries.

While the data existed, it was often “trapped” behind rigid visualizations that couldn’t anticipate every nuance of a program leader’s needs. Furthermore, the presence of personally identifiable information (PII) meant that certain qualitative details, the very context that makes data actionable, remained restricted and difficult to surface safely.

To bridge this gap, AWS developed TARA (Technical Analysis Research Agent). While TARA has been built for the internal analytics needs of AWS, the Dataset Q&A capabilities that we used are available to Quick customers facing similar challenges. Built by the Specialist Data Lens (SDL) team, TARA is an AI-powered analytics assistant that uses the custom chat agent capabilities of Quick. TARA serves as a unified conversational interface that you can use to explore multiple integrated datasets, live system APIs, and specialized research agents through natural language. By using MCP to securely connect structured datasets with external systems and domain-specific research agents, TARA bridges the gap between quantitative metrics and qualitative context. This allows leaders to tie quantitative metrics to the ground truth of what’s happening in the field, enriching analytical insights with real-time operational context while making sure sensitive PII remains protected.

We evolved TARA’s conversational analytics capabilities by adopting the Dataset Q&A feature as the foundation for semantic query generation and insight delivery. This post explores that journey and the impact of business users interacting with data more naturally. By embedding semantic definitions directly into the dataset and grounding SQL generation in the business meaning of the data, Dataset Q&A significantly improved the quality and reliability of insights. This enhancement delivered more than a 48 % improvement in response accuracy, reduced query failures to near zero, and shortened analysis time from hours to minutes.