Abstract
Business Intelligence (BI) increasingly merges dashboard interaction with LLM-based assistance, yet these modes often fall out of sync during multi-step analyses.
As users toggle between direct dashboard manipulation and natural language queries, maintaining a consistent analytical state across filters, hierarchies, metrics, and chart contexts becomes challenging. We introduce TwinBI, an agentic digital twin framework that integrates an LLM-based agent system with an executable BI dashboard state.
TwinBI unifies conversational interaction, dashboard manipulation, semantic grounding, and provenance tracking through a shared analytical state reconstructed from a unified interaction log. It exposes artifacts such as schema views, SQL, logs, and an /insights command for state-grounded analytical summaries.
We evaluate TwinBI in two complementary ways. In a controlled A/B benchmark with the same backbone agent, TwinBI improves exact-match accuracy from 43.3% to 63.3%, partial-credit accuracy from 48.3% to 70.8%, and significantly reduces timeout rate from 40.0% to 10.0% relative to Dashboard alone.
In a usability study, participants benefited from the integrated dashboard-and-chat workflow, achieving high task accuracy, moderate workload, and favorable ratings for state-aware interaction mechanisms. These results indicate that TwinBI enhances both agent-level analytical reliability and user-facing analytical support by transforming visible dashboard state into richer actionable context.
Our dataset and source code are available at: GitHub - TwinBI
Blogger's Review: TwinBI illustrates the potential of deep integration between LLMs and BI dashboards, enhancing both analytical accuracy and user interaction experience. This intelligent digital twin design concept may pave the way for new directions in future BI tools, making it a noteworthy development to watch closely.