From Numbers to Narratives: A Generative Artificial Intelligence Framework for Automated Risk Report Production in IT Program Portfolio Governance

Abstract

Every IT program portfolio generates a continuous stream of quantitative performance data. What it rarely generates, and what governance institutions consistently demand, is a coherent written account of what those numbers mean, what risks they signal, and what program boards should do about them. Human risk analysts currently bridge this gap, translating earned value metrics and delay risk flags into governance narratives manually, at considerable cost, with variable quality, and at a pace that often lags the reporting cycle. This paper introduces PRISM, the Predictive Risk Intelligence with Synthesized Management narratives framework, a generative artificial intelligence pipeline that automates the production of professional, governance-ready risk narrative reports by combining XGBoost delay risk classification, TreeSHAP feature attribution, and a fine-tuned large language model trained on a curated corpus of 2,340 authentic program risk reports drawn from federal IT investment governance archives. The framework accepts a standard program performance data vector as input and produces a complete, contextually accurate, and institutionally appropriate governance risk narrative as output, indistinguishable in professional quality from reports produced by experienced human risk analysts. Evaluation on a held-out test set of 412 investment records demonstrates that PRISM achieves a BLEU-4 score of 0.743, a BERTScore F1 of 0.891, and a practitioner Turing evaluation pass rate of 74.6 percent, meaning that in nearly three out of four cases, certified program management practitioners rated PRISM-generated narratives as of equal or superior quality to human-authored equivalents without being able to identify their machine origin. A governance decision quality study with 56 active program board members confirms that decisions made using PRISM narratives are 31.4 percent more likely to target the correct primary risk driver than decisions made using raw performance dashboards, and that PRISM narratives reduce mean risk report production time from 4.2 hours to 11 minutes per investment without loss of governance utility. These findings establish generative AI narrative synthesis as a technically feasible, practically superior, and institutionally deployable alternative to manual risk report production in IT program portfolio governance contexts.

Full text

Cite this publication

Nkem Daniel Obaloje, Ezeokechukwu Chiemere Victor (2026). From Numbers to Narratives: A Generative Artificial Intelligence Framework for Automated Risk Report Production in IT Program Portfolio Governance. International Journal of Engineering and Computer Science, 15(07), 28669-28677. https://doi.org/10.18535/ijecs.v15i07.5564

Related publications

• Cognitive Supplier Risk Management Systems for Medical Device: AI Framework for Predictive Quality, Compliance, and Supply Chain Resilience

  International Journal of Engineering and Computer Science · 2026

• Privacy-Preserving Synthetic Data for HIV Risk Prediction in People Who Inject Drugs in Kenya: Benchmarking Differentially Private Generative Models

  International Journal of Engineering and Computer Science · 2026

• AI-Assisted Tasawuf Education: Developing the AI-Murshid Framework

  International Journal of Scientific Research and Management (IJSRM) · 2026

• Short‐Term Profits for Long‐Term Value: Corporate ESG Dilemmas Under Earnings Pressure and Pathways to Governance

  Accounting & Finance · 2026

• Foundation Models for Medical Imaging: Large-Scale Pretraining for Universal Diagnostic Systems

  International Journal of Health and Medical Research · 2026