Artificial intelligence (AI) is the buzzword of the 21st century. The term is thought to originate from a 1956 project proposal by computer scientist John McCarthy, in which he asked the Rockefeller Foundation for financial support for a scientific conference. But, until the past decade, technical limitations restricted the field.
Recently, processors have become faster, and larger quantities of digitised data have become available. As a result, learning algorithms have been implemented in the financial services industry. An interplay of AI, data analytics and cyber security is creating cross-industry platforms that connect people, machines and processes.
One possible application of AI that has arisen is for the prediction of cashflows between bank branches. Companies and banks need their revenues to arrive as quickly as possible in their accounts – in near real time, if possible. One method deployed is a ‘safe solution’, where the daily turnover in cash is transferred automatically into bank accounts. Specialists then verify whether the balance of the account and the amount of money in the safe match.
The analysis of daily turnover for each specific bank branch is important but takes significant manpower and resources. An AI solution could improve this process through the reduction of time and effort, combined with increased security. Targens, a software company for banks and financial institutions, was given the task by one of its clients – Cash Logistik Security – to develop a prototype of an AI solution. Cash Logistik Security, founded in 1998, provides cash management services to financial institutions throughout Germany.
“The aim was to achieve a precise prediction of the daily turnovers for each branch,” says Ines Planner, a data scientist for Targens. “Based on this information, turnovers that are not transmitted correctly to the accounts can easily be discovered.”
As a result, specialists only have to examine the anomalies where turnover differs significantly from predicted values. Meanwhile, security is strengthened because AI highlights abnormalities the human eye may overlook.
To develop the software, true values on turnover were given to Targens to conduct supervised learning with the AI system. The data was split into two datasets, one each for training and testing; the former was used to train the model and the latter to analyse the model’s performance. Nowcasting techniques were then applied, with turnover at a particular branch predicted using activity at other branches.
“Classical and nowcasting models both have their specific advantages,” says Planner. “The idea was to combine the different models to one hybrid model for each branch. Depending on specific features, the hybrid solution puts more weight to one model.”
Accurate forecasting is crucial for banks and for central banks. Risks need to be calculated and suitable measures need to be implemented. Targens’s hybrid solution fulfils these requirements.