Scholarlink

<p>The interpretation of complex financial data often poses significant challenges for small and medium-sized enterprises (SMEs), as traditional reporting systems typically remain static. This paper presents a&#xa0;controlled multi-agent architecture for a&#xa0;dialog-based AI financial assistant designed to address this gap. The technological core of the approach lies in systematic context engineering, enabling a&#xa0;strict functional separation between deterministic data acquisition (retrieval agent) and narrative interpretation (explanation agent). Through the coordination of a&#xa0;central orchestrator and targeted memory management, the system ensures that AI-generated responses are consistently grounded in a&#xa0;verified single source of truth, thereby minimizing the risk of misinterpretation or incorrect statements. Based on these architectural decisions, we derive six transferable design principles that condense key design choices for controlled conversational financial analytics. Using a&#xa0;prototypical cash flow analysis as an exemplary case, the paper demonstrates how this orchestrated system transforms financial metrics into accessible, narrative-driven financial analyses. The results of a&#xa0;user-centered evaluation suggest that this approach reduces the perceived barrier to engaging with complex analyses. In doing so, the work provides a&#xa0;practical contribution toward making financial expertise more broadly accessible through modern agentic AI.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Dialogbasierte Finanzanalyse für KMU: Agentic AI zur kontrollierten Interpretation komplexer Kennzahlen

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月3日

查看全文
  • Miriam Velasco,
  • Simone Braun

摘要

The interpretation of complex financial data often poses significant challenges for small and medium-sized enterprises (SMEs), as traditional reporting systems typically remain static. This paper presents a controlled multi-agent architecture for a dialog-based AI financial assistant designed to address this gap. The technological core of the approach lies in systematic context engineering, enabling a strict functional separation between deterministic data acquisition (retrieval agent) and narrative interpretation (explanation agent). Through the coordination of a central orchestrator and targeted memory management, the system ensures that AI-generated responses are consistently grounded in a verified single source of truth, thereby minimizing the risk of misinterpretation or incorrect statements. Based on these architectural decisions, we derive six transferable design principles that condense key design choices for controlled conversational financial analytics. Using a prototypical cash flow analysis as an exemplary case, the paper demonstrates how this orchestrated system transforms financial metrics into accessible, narrative-driven financial analyses. The results of a user-centered evaluation suggest that this approach reduces the perceived barrier to engaging with complex analyses. In doing so, the work provides a practical contribution toward making financial expertise more broadly accessible through modern agentic AI.