Cash Flow Optimization in ATM Network System

Reduce operating costs for providing cash to ATMs.

To analyse the data, we used real daily cash withdrawal data from an ATM. We utilized Gradient Boosting Regressor to build an ML-model.
The solution comprised three stages.
The first stage consisted of:

• data evaluation
• requirements and success criteria determination
• data loading, depersonalization and enrichment
• agreement on the experiment procedure

The second stage focused on:

• segmentation of the research objects;
• training, testing and evaluating the quality of the model.

The third stage included:

• automated data loading or model deployment in the customer’s environments;
• regular quality control through A/B testing;
• technical support of the model and optimization of new data entry.

We implemented automatic forecasting of cash demand with an error of 0.01-3.5%.
The bank reduced operating expenses: the amount of allocated funds by up to 30%, cashback by up to 40%, and out-of-cash downtime by up to 0.2%.