Data-Driven Cash Replenishment Planning of Recycling ATMs with Out-of-Cash and Full-of-Cash Risks

Introduction

Nowadays, many banks outsource the cash management operations linked to automated teller machines (ATMs) to cash-in-transit (CIT) companies. CIT companies not only determine ATM replenishment plans but also implement these plans with their resources such as armored trucks and staff. Every day, trucks deliver cash cassettes from a depot to selected ATMs, and cassettes in the ATMs are exchanged for new ones. Traditional ATMs provide only cash withdrawal service, while the newer recycling ATMs allow both cash withdrawal and cash deposit to be made. Currently, Recycling ATMs are more commonly used in China. Decisions about whether any given ATM should be replenished and by what amount of cash are an important but difficult task for CIT companies. The ATM replenishment plan will determine the costs (including the operational cost, reflected by the number of replenishment tasks, and the opportunity cost, reflected by the average cash inventory in ATMs) and the service level (reflected by the rates of out-of-cash and full-of-cash occurrence).

Traditionally, ATM cash usage forecasts and replenishment plans have been determined manually, generally simply according to experience, but this is typically both costly and inefficient (Simutis et al., 2007; Armenise et al., 2010). Adequate service level is an important objective for companies, so the replenishment decisions should be formulated without compromising the service level (Sivakumar et al., 2013). The demand for cash fluctuates across locations, seasons, customers and so on. Furthermore, the increasing use of recycling ATMs significantly increases the difficulty of forecasting cash requirements. Therefore, many CIT companies are turning their attention to more efficient management of their cash in ATM networks (Westland, 2002; Drehmann et al., 2002).

The ATM replenishment problem is similar to vendor-managed inventory (VMI, Hohmann & Zelewski, 2011), in that a CIT company needs to determine which ATMs are to be replenished and the amount of cash for each. The most common inventory models include the periodic model (Gürler et al., 2014), the (s, S) model (Janssen et al., 2018), and the (r, Q) model (Alavi et al., 2016). In the periodic model, the inventory is reviewed and replenished at fixed intervals. In the (s, S) model, the level of inventory is replenished to S whenever it drops to s. In the (r, Q) model, the inventory is replenished by quantity Q whenever the level of inventory drops to r. The safety stock and replenishment quantity are usually calculated by the safety stock method (Schmidt et al., 2012) and economic order quantity (EOQ) model (Harris, 1913; Pentico et al., 2014) respectively.

In this paper, the authors formulate ATM replenishment as a special inventory problem on the basis of the (s, S) model. Unlike previously proposed inventory models, two safety stocks are established for each ATM, corresponding to out-of-stock risk and full-of-stock risk, which is better suited to the analysis of recycling ATMs. A model is established to forecast the cash usage during the replenishment lead time (usually one day). Both the safety stocks and the replenishment quantities are calculated on the basis of a common service level for all ATMs, using the percent point function constructed by the cash usage forecast and the variance in historical cash usage. If the amount of cash remaining is less than the lower safety stock or larger than the higher safety stock, then the ATM is visited, and cash will be replenished to a certain level. By setting the common risk parameters, a common service level and an appropriate replenishment frequency can be planned for all ATMs. More importantly, the computational cost (in terms of computational time) of the model is quite low, which will facilitate its use in industry. Experiments with real data are conducted to examine the effectiveness and efficiency of the established model.

The remainder of this paper focuses on the related literature review, an appropriate forecasting model, cash replenishment model, experiments based on real-world data, results of a sensitivity analysis of parameters and conclusions.