This retrospective study described the development of an AI model that could identify key preoperative variables associated with failure to rescue (FTR) after coronary artery bypass grafting (CABG). FTR was defined as postoperative mortality within 30 days due to stroke, renal failure, reoperation, or prolonged ventilation. 9,974 patients from the STS institutional cardiac surgical base were selected, with an overall FTR rate of 2.5% among the population. Initially, 35 peroperative variables were selected based on prior literature, studies, and author expertise. Researchers used the RFECV model, a greedy search feature selection algorithm, to narrow down these variables to only key risk factors. For each variable, the model assigned a SHAP value, which indicated how much impact each variable had on the outcome (FTR). Of those variables, the model identified 12 key risk factors, with the most influential factors being low levels of hematocrit, ejection fraction, creatinine, and platelet count, as well as high levels of bilirubin, BMI, WBC count, age, hemoglobin A1c, and a high MELD score. Additionally, the model evaluated the effects risk factors had on each other; for example, high albumin and high hemoglobin levels were associated with strong positive P values, while low values of both caused more variability in prediction. The model’s predictive performance was analyzed using a random forest model, with the best precision-recall curve, calibration slope, and brier score reflecting the model’s relatively high predictive ability (0.78, 0.82, and 0.02, respectively).