From Data to Insight Without Being Asked – Insightful AI for Business Data

In enterprise analytics, most LLM-based tools merely retrieve facts or summarize charts. But insights that truly drive strategy—those rooted in domain expertise—still elude most automated systems.

Our paper, Data-to-Dashboard, presents a novel agentic AI framework that simulates how human analysts think. By combining domain detection, concept extraction, multi-perspective analysis, and iterative reflection, this system transforms raw tables into dashboards that don’t just display data or answer questions—they tell the right story. Unlike previous chart QA methods, it doesn’t rely on question templates or closed schemas. It reasons.

Evaluated on diverse business datasets, our approach outperformed prompt-only GPT-4 baselines in insightfulness, novelty, and depth. We believe this opens up a new direction in business analytics—one where AI doesn’t wait to be asked, but actively surfaces what matters.