Automated Sales Lead Scoring and Prioritization: A CRISP-DM Comparison of Four Machine Learning Classifiers

Abstract— Sales teams that receive more leads through Customer Relationship Management (CRM) channels than they can engage often invest follow-up efforts on low-probability leads while more promising ones end up neglected. In practice, prioritization continues to depend on manual judgment or established fixed-point rules, and previous predictive studies vary in their algorithm choices, utilize different datasets and metrics, and rarely address overfitting or translate predictions into actionable priorities, necessitating a reproducible and implementable scoring method. This research develops an automated model for lead scoring and prioritization based on the CRISP-DM framework.Secondary data were retrieved by the authors on 21 October 2026 from an open lead scoring dataset hosted on GitHub, and all 9,240 records and 37 attributes were used without additional sampling, followed by a stratified 80:20 train–test split. Logistic Regression, Decision Tree, Random Forest, and XGBoost were compared using stratified five-fold cross-validation, leakage control, and accuracy, precision, recall, F1-score, ROC-AUC, and lift. XGBoost achieved the highest mean ROC-AUC (0.9852) and F1-score (0.9318), while Logistic Regression (0.9818) and Random Forest (0.9816) were close behind, so interpretable and ensemble models are both viable. The best model's probability is converted into a 0–100 score and three priority levels and delivered through a proposed system architecture, workflow, and mobile prototype. The study concludes that probability-based scoring can replace manual prioritisation, and it recommends validation on organisation-specific CRM data.