Explainable Machine Learning for Evidence-Based Management Decision-Making and Organizational Performance Forecasting

Organizations turn to machine learning for strategic insights, transparent predictions can enhance trust and accountability for management. This study created explainable models to predict future firm growth and pinpoint executive and strategic factors related to firm performance for the coming year. The analysis was performed on 188 firm-year observations from 22 NASDAQ-listed companies over the last ten years (2006-2015), with 144 observations used for training and 44 observations for testing. Ordinary least squares, Elastic Net, Random Forest and Gradient Boosting were compared to historical-mean and persistence benchmarks. The best model was used for model-agnostic permutation importance, which was used to interpret the model. Gradient Boosting outperformed the other models in test performance, with an MAE of 20.038 and RMSE of 29.300, and R² of 0.105. The most important factors for predictive accuracy were previous growth, estimated growth and executive functional background; very little incremental information was added by most governance and strategic-posture variables. The results suggest that explainable forecasts may serve as a tool for managers to reflect on their performance expectations, uncover uncertainty, and question assumptions, but not as causal evidence. Combined with strategic theory and contextual knowledge, and with accountability in managerial judgment, explainable machine learning can provide useful support to decision-making.