Book

From Numbers to Narratives: A Generative AI Framework for Automated Risk Report Production in IT Program Portfolio Governance Every IT program portfolio produces mountains of performance data, but almost none of it becomes the clear written narrative that governance boards actually need. Analysts currently bridge that gap by hand, at real cost in time and consistency. This paper introduces PRISM, a generative AI pipeline that closes the gap automatically, combining XGBoost delay risk classification, SHAP based feature attribution, and a fine tuned language model trained on thousands of authentic program risk reports. Given a standard performance data vector, PRISM produces a complete governance ready risk narrative that reads as though it came from an experienced analyst. Tested against 412 held out investment records, PRISM narratives were rated equal or superior to human authored reports nearly three out of four times by certified practitioners, unable to tell which was machine generated. A companion study with 56 program board members found that decisions made using PRISM narratives were far more likely to target the true root cause of risk than decisions made from raw dashboards alone, while cutting report production time from over four hours to about eleven minutes. Together these results position generative narrative synthesis as a practical, deployable upgrade to how IT portfolio governance actually works.