We present a predictive-prescriptive analytics framework that translates hourly precipitation forecasts into operational early warning triggers for Andean regions. The framework was evaluated using data from six hydrometeorological stations in the province of Pichincha. Furthermore, four models were trained and compared using the following metrics: mean absolute error, root mean square error and R2. The models Prophet, Random Forest, XGBoost and LSTM were trained and compared using these metrics. Taking the mean absolute error into account, the results indicate that Random Forest systematically obtained the lowest errors: 0.070 in Chalpi Grande, 0.074 in San José de Minas, 0.079 in Quijos, 0.091 in San Francisco, 0.126 in El Carmen, and 0.161 in Blanco Chico. XGBoost performed competitively, achieving the second lowest error at most stations: 0.073 in Chalpi Grande, 0.076 in San José de Minas, 0.078 in Quijos, 0.096 in San Francisco, 0.158 in El Carmen, and 0.163 in Blanco Chico. LSTM produced higher MAEs, with values of 0.131 in Quijos and 0.287 in Blanco Chico. Prophet produced the highest errors: 0.605 in San Francisco and 0.870 in Chalpi Grande. Furthermore, intensity thresholds were derived from the predicted distributions for each station to prescribe actions, with reference values of 15.25 mm/h in San Francisco, 18.4 mm/h in El Carmen, 12.8 mm/h in San José de Minas, and 10 mm/h in Blanco Chico Alto. This framework increases operational utility by combining probabilistic prediction with decision-making under risk, while providing transparent, reproducible and adaptable triggers for local early warning systems.