Data-to-Dashboard: Multi-Agent LLM Framework for Insightful Visualization in Enterprise Analytics

TL;DR

This work proposes a two-stage, agent-based framework that unifies data-to-insight and insight-to-chart processes for enterprise analytics, grounding analyses in domain knowledge without fixed ontologies. Stage 1 profiles data and uses domain and concept detectors to produce descriptive, predictive, and domain-oriented insights, refined through iterative self-reflection. Stage 2 applies Tree-of-Thought reasoning and a three-expert consensus to select visualization strategies that preserve domain significance and narrative. Across multiple datasets, the approach improves insightfulness, novelty, and depth compared to baselines, and demonstrates alignment with ground-truth directions while enabling human-in-the-loop validation for practical deployment.

Abstract

The rapid advancement of LLMs has led to the creation of diverse agentic systems in data analysis, utilizing LLMs' capabilities to improve insight generation and visualization. In this paper, we present an agentic system that automates the data-to-dashboard pipeline through modular LLM agents capable of domain detection, concept extraction, multi-perspective analysis generation, and iterative self-reflection. Unlike existing chart QA systems, our framework simulates the analytical reasoning process of business analysts by retrieving domain-relevant knowledge and adapting to diverse datasets without relying on closed ontologies or question templates. We evaluate our system on three datasets across different domains. Benchmarked against GPT-4o with a single-prompt baseline, our approach shows improved insightfulness, domain relevance, and analytical depth, as measured by tailored evaluation metrics and qualitative human assessment. This work contributes a novel modular pipeline to bridge the path from raw data to visualization, and opens new opportunities for human-in-the-loop validation by domain experts in business analytics. All code can be found here: https://github.com/77luvC/D2D_Data2Dashboard

Figures (5)

• Figure 1: Existing approaches, whether agentic or non-agentic, use language models to obtain context-specific answers and insights, often overlooking the deeper value still embedded in the underlying raw data.
• Figure 2: Our end-to-end data-insight-visualization approach provides context-independent domain-aware insights, thus overcoming the limitations of existing systems.
• Figure 3: This figure compares the insights obtained with and without domain identification. As can be seen, domain identification grounds the resulting insights, fending it from hallucinations and providing the business analyst with more confidence.
• Figure 4: Comparison of our generated insights with InsightBench ground truth. Our system captures the core analytical direction, identifying key themes such as cost variability, processing dynamics, and optimization opportunities, with broader concept coverage. However, it misses one specific ground truth angle—variation across users and departments—highlighting an area for improvement in capturing organizational structure–related insights
• Figure 5: Examples of insightful figures generated by our approach