ATM supply scheduling optimization: From demand prediction to intelligent route planning

Currently, automated teller machines (ATMs) are one of the main sources of cash circulation in societies. Banks and their clients depend heavily on their availability and ability to serve client requests consistently. To achieve this high availability, banks are trying to optimize their ATM resupply planning, ensuring client satisfaction while minimizing their own costs and the amount of capital that remains idle inside ATMs. The solution proposed in this paper attempts to create a unified pipeline that predicts the optimal time for an ATM to be re-supplied using machine learning techniques such as Linear Regression and Deep Learning, and then optimizing the re-supply planning using common optimization algorithms like Integer Linear Programming. The algorithms consider historical data and external data that may affect the resupply process or client activity per ATM, such as holidays or street events near the ATM. The route optimization algorithms find the optimal supply planning, minimizing costs and energy consumption. The presented platform was evaluated on real world dataset provided by the National Bank of Greece, proving both the feasibility and the effectiveness of the platform in real world banking operations. The results of the evaluation show that the proposed solution manages both to predict the optimal re-supply time for most ATMs and find the optimal route planning for the proposed ATMs, covering the needs of the bank and minimizing its costs while maximizing customer satisfaction. The innovation of the presented solution is a holistic approach that creates a full optimization pipeline, deciding which ATMs should be re-supplied each day and then optimizing the supply plans for the selected ATMs. The impact of the presented work is invaluable both in research and financial sectors since it optimizes the daily logistics of banks, which are cornerstones of modern cities.