The customer problem: The bank faced significant challenges in managing cash reserves across its network of 490+ ATMs and multiple vaults. Excess cash storage, known as "dead cash," led to inefficiencies and unnecessary costs. The goal was to minimize dead cash while ensuring optimal cash availability for customer transactions.
Solution: To tackle this issue, a mult-tiered approach was used. A neural network model (ML model) was used to forecast the cash demand at each ATM location. The model analyzed past usage patterns and change request trends to predict future cash needs with an accuracy exceeding 85%. This input was used in an OR model to determine the minimal amount of funds required to support each ATM node based on the currency denominations to be disbursed.
Implementation and Results:
Forecasting Accuracy: The resulting models’ high accuracy in predicting cash requirements allowed the bank to streamline cash loading schedules and reduce excess reserves.
Dead Cash Reduction: The implementation of the solution resulted in a 47% reduction in dead cash across the bank's ATM network, significantly improving cash flow management.
Enhanced User Experience: By optimizing cash loading requirements, the bank improved customer experience, ensuring that ATMs were consistently stocked with the appropriate denominations for their target locale.
Rapid ROI: The solution proved to be highly effective, achieving a return on investment (ROI) within a few weeks of deployment.
Recognition: The solution's success led to its productization, and it was subsequently awarded a national software quality award, recognizing its innovative approach and significant impact on operational efficiency.
