This research addresses service-level inefficiencies in the dark kitchen sector, a rapidly growing business model within the food delivery industry. The analyzed operation initially exhibited a service level of 78.2%, significantly below the target service level for last-mile delivery operations, highlighting the need for operational improvement during peak demand periods. To address this gap, an integrated improvement model was developed and validated through a before-and-after pilot study conducted under real operating conditions. The proposed model combines three complementary engineering tools: (i) a Machine Learning–based decision-support system for delivery routing, (ii) a Linear Programming model for optimal staff allocation during peak hours, and (iii) process standardization through digital Standard Work in the packing verification stage. Performance was evaluated using key indicators such as on-time delivery, order integrity, and labor productivity. The results show a 15.2% improvement in on-time delivery rate (from 79% to 91%), a 64% reduction in packing error rate (from 11.8% to 4.2%), and a 36.8% increase in staff productivity (from 12.5 to 17.1 orders per staff-hour), leading to an overall OTIF improvement from 78% to 91%. The main contribution of this study lies in the integrated application of predictive analytics, mathematical optimization, and process standardization within a single operational framework, providing a practical and replicable solution for dark kitchens and similar service-based food delivery operations facing high demand variability.