A discovery framework for AI innovation projects

Abstract

Traditional Requirements Engineering (RE) practices face limitations when dealing with the development of data- and model-centric Artificial Intelligence (AI) systems. The literature reveals a lack of methodological support for problem discovery activities in AI innovation projects, which compromises subsequent development phases. To mitigate this issue, this work introduces DIP-AI, a discovery framework designed to guide early-stage exploration in such initiatives. Grounded in a tertiary study, the framework harmonizes technical processes from the ISO/IEC/IEEE 12207 and ISO/IEC 5338 standards with the agile Design Thinking approach, while also incorporating 5W2H and GUT matrices for risk management and prioritization. The proposal was validated through a multi-method strategy: (i) an empirical case study within an industry-academia project; (ii) a controlled evaluation involving graduate students in Software Engineering and AI; and (iii) semistructured interviews with innovation project coordinators. Although the results confirm the utility and strategic value of DIP-AI in mitigating risks and aligning stakeholders, they also highlight barriers related to cognitive overload when using the framework. Consequently, an assistive tool based on Large Language Models (LLMs) was developed and qualitatively evaluated via a focus group with industry professionals. The results demonstrate that the integrated framework-tool approach reduces initial methodological friction, accelerates the learning curve, and fosters a strong intention for continued adoption. As an academic contribution, this research fills a methodological gap in the RE for AI literature by providing a rigorous and replicable conceptual model. For the industry, this deliverable establishes itself as a relevant and practical asset, enabling organizations and multidisciplinary teams to structure the discovery of complex AI problems with greater predictability, thereby reducing resource waste and maximizing product success and sustainability.