New product development (NPD) failure rates range from 40% (Cooper, 2019) to 90% (Kim et al., 2016), so firms must allocate resources effectively, including abandoning failing projects if necessary. However, when to abandon is often unclear, so portfolio managers risk commitment escalation (Eliens et al., 2018) and overinvesting. This study explored the research question: How should firms evaluate whether to continue or abandon projects in high-tech R&D? A dual approach was taken: First, to gain foundational insights, an innovative AI-assisted literature review method was developed and used to create a novel R&D decision-making framework. Second, to test the key concepts of the framework and illustrate its application, an agent-based simulation (Sulis & Taveter, 2022) compared the performance of the framework against risk-based and ROI-based approaches in a 1000-project portfolio. The framework outperformed the other strategies in simulations and emphasised the importance of gatekeeper independence, negotiating clear criteria up front and using formal decision gates. The simulation results also suggested actionable guidance for practitioners: For a feasibility phase to add value, it should cost less than 30% of the project budget and significantly improve cost estimates. Theoretical contribution: This study integrated key NPD decision theories from real options, behavioural decision-making, and portfolio management and developed a novel decision-making framework while exemplifying the use of AI to enable rapid, scalable research. It then demonstrated that decision gates outperform ROI- and risk-based approaches in simulation. Practical contribution: For R&D portfolio managers, the study provides a structured decision-making approach and guidance for sizing feasibility studies. For researchers, it provides a methodology for accelerating and scaling literature reviews.