Budget versus actual analysis: an approach to explainable artificial intelligence in finance planning and analysis

• Ano de defesa: 2025
• Autor(a) principal: Penna, Felipe Tempone Cardoso
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Análise de variações; Bayesian models; Explainable Artificial Intelligence (XAI); Financial Planning and Analysis (FP&A); Inteligência Artificial Explicável (XAI); Modelos bayesianos; Planejamento e Análise Financeira (FP&A); Variance analysis
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/12/12142/tde-24032026-171418/

This study investigates the application of Explainable Artificial Intelligence (XAI) within Financial Planning and Analysis (FP&A), focusing on Budget versus Actual (BvA) analysis. Building upon a case study of an in-house developed analytics tool, the research adopts a mixed-methods design that combines Design Science Research Methodology (DSRM) with qualitative inquiry. The first stage reports on the design and deployment of a BvA Analytics model developed in a multinational healthcare company, highlighting its ability to automate variance detection, increase coverage of P&L lines, and enhance the efficiency of financial closings. The model is then extended through a Bayesian probabilistic layer, integrating anomaly weighting and explainability features to strengthen decision-support capabilities. To evaluate and validate the solution, semi-structured interviews were conducted with senior finance leaders from multinational organizations across diverse industries. The interviews confirmed the solutions relevance, applicability, and usability, particularly in complex, multi-entity environments, while also identifying contextual boundaries where its incremental value may be limited. Feedback further refined the model by emphasizing the importance of transparent outputs and alignment with managerial judgment. The findings contribute to both practice and theory by demonstrating how traditional BvA methodshorizontal and vertical analysisremain essential, while AI and XAI enhancements provide automation, scalability, and trust. Ultimately, this research bridges the rigorrelevance gap in FP&A by delivering an open, replicable approach to financial variance analysis, validated by expert practitioners, and offering a pathway for future adoption and comparative studies.