Proposta de processo para identificação e priorização inicial de riscos em projetos apoiado por inteligência artificial generativa

Abstract

The initial identification and prioritization of risks is a project management stage that depends on the analysis of information distributed across documents, communications, and prior decisions. Generative artificial intelligence models have begun to be used to support this work; however, their use still requires an understanding of reliability, explainability, and operational limitations. This study aimed to develop and conceptually validate a process for the initial identification and prioritization of project risks supported by generative artificial intelligence. The research adopted a qualitative, applied, and exploratory approach, guided by Design Science Research and organized into four phases: a systematic literature review, process design, conceptual validation with fifteen project managers, and final refinement. The validation assessed the artifact across seven dimensions (usefulness, ease of use, applicability, reliability, explainability, process quality, and quality of the generated output) through semi-structured interviews and deductive thematic analysis. The results indicate that generative artificial intelligence supports the initial identification of risks by synthesizing information and producing broader preliminary lists, provided that it is accompanied by quality documental inputs, clear contextualization, and mandatory human review. The central finding showed that the applicability of the process depends on the existence and quality of the project's documental artifacts, and that managers' trust derives less from the model's technical performance and more from the traceability of information, the clarity of the steps, and the preservation of human control over the decision. Fourteen of the fifteen managers expressed intention to use the process. The refinement produced version 2.0 of the process, comprising eleven steps, a versioned structured prompt, output and record templates, a human validation checklist, and an implementation manual. The study delivers a structured and empirically validated process for the controlled use of generative artificial intelligence in the initial identification and prioritization of risks in project contexts.