Techniques are provided that may include receiving sensor data from one or more machine sensors, the sensor data indicating machine failure. Informational data about current repair technician backlog, shop availability backlog, and parts availability are retrieved and inputted to a machine learning model for computing an estimated time to when the failed machine will be repaired. Customer data and dealer site data are retrieved and an estimated financial impact from a potential loaner machine is computed. The various computed and retrieved data are inputted to a decision tree algorithm, which computes an optimal solution for maintaining customer productivity. The network-accessible dealer site is notified of the optimal solution, which is intended for the customer.