Artificial Intelligence in Banking Sector
One of the most inevitable effects of recent advances in modern sciences and technologies is the rapid increase in both dimensions, and complexity of the problem and data requirement. This phenomenon motivates problem solvers to develop more sophisticated approaches. Modern algorithms can now address this challenge by utilizing intelligence in their problem-solving processes.
Intelligence can be defined as making the right decision, given a set of inputs and various possible actions. In problem solving, this is referred to as Artificial Intelligence (AI), and is achieved by systematizing intellectual tasks related to human intellectual activity. We employ our intelligence to adapt ourselves to the different situations we encounter; in other words, we learn the best strategies in each situation, and store them in our brain to recall when similar situations arise. Learning, according to David Fogel, is an intelligent process in which the basic unit of mutability is the idea. “Good” adaptive ideas are maintained, much as good genes increase in a population, while poor ideas are forgotten. In insect societies, this only requires the evaporation of pheromone trails, while in humans it requires time to forget. According to Albert Einstein, the measure of intelligence is the ability to change.
Alan Turing started the artificial intelligence approach when he worked out how mental computations could be broken down into a sequence of steps that could be mechanically simulated. In the 1950s, when the term “artificial intelligence” had not yet been introduced, Alan Turing came up with the idea of building intelligent machines. He proposed that if a problem could be expressed as an algorithm, or a precise set of formal instructions, then it could be computed mechanically by a machine. Turing proposed that if this machine’s responses were indistinguishable from a human’s, then the computer could be considered a thinking machine, and this became known as the Turing Test. The Turing machine is one of the most important breakthroughs of the 20th century that led to both the invention of the modern computer, and new ways of thinking regarding human cognition.
In the optimization and data science domains, we have such algorithms whose power comes from their ability to learn new features as they work on high quality solutions for large scale problems. These intelligent approaches can learn, memorize, and maintain a single candidate solution or a population of solutions that provides the information acquired by the process, as well as form the basis for making future decisions. The use of prior knowledge to create adapted solutions can sometimes be interesting, innovative, and even competitive against human expertise. Most researchers accept intelligence as an umbrella that covers intellectual activities.
Since the time Alan Turing created the Turing machine in 1950 and John McCarthy named this approach artificial intelligence at a conference in New Hampshire in 1956, the focus of artificial intelligence has changed from creating a robot as intelligent as a human, to having algorithms that can learn in a way similar to the human brain while solving problems. Intelligence is among the striking features that draw the line between human excellence and that of other beings, and now the humans are revealing the difference between the powerful algorithms and others.
AI has found another application in the banking sector; forecasting automated teller machine (ATM) cash demands. Although online banking has been increasing remarkably among customers, ATMs are still a critical part of their daily financial routines. Since installation of the world's first ATM by Barclays Bank in London in 1967, the number of ATMs has already reached over 3million worldwide.
The future trends of ATMs clearly reveal the critical requirements for managing ATM networks in terms of both providing high quality service, and minimizing overall costs; in other words, optimizing operations and maximizing profits in ATM networks. The ATM cash optimization problem is basically a two-phase problem: forecasting cash demands for each ATM, and optimization of ATM-related activities. While in the forecasting phase accurate forecasts of withdrawal patterns for each ATM are generated by employing prediction algorithms, in the optimization phase a replenishment schedule is produced for the entire ATM network.
Because of the continuous nature of the predicted amounts, several prediction algorithms are proposed for literature, and for the business cases. One of the most known algorithms, linear regression, can model linear dependency between predictors and the label variable. However, AI algorithms, like artificial neural networks (ANN), have non-linear mapping and strong self-learning capabilities. Each AI algorithm can detect different types of information about the problem instance, and present different performances under various circumstances. While one algorithm is more accurate for forecasting cash-in for one ATM, another approach can produce more accurate results for another ATM. To boost each of their strengths, algorithms can also be used in hybrid or ensemble learning forms, whose abilities to improve accuracy in state of the art industry applications are already proven.
As we celebrate the ATM’s 50th birthday, the ATM seems to continue being a primary instrument in the future of individual payments. And by means of AI algorithms, we can offer an effective solution which can create up to 30% savings on ATM-related costs for retail banking institutions, including operations, transfer, and interest costs.