The Use of AI in Banking – Part II

The first part of the article explained the added value of using AI, including the use of intelligent chatbots, Personal Voice Assistants (PVA) and marketing Use Cases as examples. The following sections discuss further areas of application for AI in the areas of credit assessment and fraud prevention, investment advice / investment, compliance and Intelligent Robotic Process Automation.

Credit assessment and fraud prevention

Risk information such as credit rating information or score values are among the most important instruments in the banking business. Credit agencies aggregate data on the payment behaviour and creditworthiness of consumers and make this data available to inquiring companies in the form of creditworthiness information. In addition to classic products such as credit rating information and scoring, the market-leading credit agencies are increasingly offering other Big Data-based risk management tools. For example, device, identity or account number checks are offered to prevent fraud in e-commerce [see SCHUFA (2018)].

Big Data’s credit rating in conjunction with AI offers new possibilities. For example, the company Kreditech analyzes up to 20,000 data points about its users, ranging from socio-demographic information about surfing behavior in social networks to installed fonts on the user’s PC and the length of time it takes to complete a credit application on the websites of Kreditech’s subsidiaries. Deep learning methods for profile building enable the creation of creditworthiness information in real time.

With AI-based credit assessment, there is a risk of discernment for the customer. Incorrect assumptions, data quality deficiencies, exceeding the model limits or improper use of the model can quickly have negative consequences for the customer. As a consequence, certain customer groups may find it difficult to access traditional bank loans or may only be offered them at comparatively higher costs. Irrespective of this, the risk of such discrimination also exists in the case of manual credit application checks. In an automated credit application, however, it is not possible to describe circumstances to the bank advisor that could lead to the credit being granted despite contrary indicative parameters.

Automated credit decisions using AI mainly relate to retail business. This means that model errors have a faster or comparatively stronger effect than human error decisions. These unwanted credit decisions have a negative impact on the portfolio.


In the money laundering suspicion check, an analysis of structured amounts of data from the customer’s transactions for behavioural problems takes place. With the help of AI methods, previously unrecognised patterns such as number anomalies and regular transactions below the threshold values can be recognised and the effectiveness of existing recognition rules can be increased by feedback loops. AI techniques are also suitable for detecting further compliance violations, e.g. unusual entries in the bank’s chart of accounts. In summary, AI-supported procedures increase the hit rate and simultaneously reduce the number of suspicious cases to be checked manually, which ultimately leads to a reduction in costs.

Investment advice / asset investment

AI can also support the development of investment strategies in the area of investment advice and investment. From historical data, e.g. price time series, company or valuation ratios, certain patterns can be identified with AI methods and concrete investment decisions can be derived from these under specification of framework conditions for the respective current market scenario.

One application example is premium strategy with Artificial Intelligence (AI) from Wallrich Wolf Asset Management AG [see Wallrich Wolf (2018)]. The system used compares the current prices of the Euro Stoxx 50, the volatility and the prices of the put options on the above-mentioned index traded on the Eurex futures exchange with those of the past. The investment rules previously optimized using intelligent computer algorithms are then applied to the patterns recognized using AI methods.

Banks have a comprehensive overview of the transaction data of all customers in their portfolio and can use AI techniques to make greater use of this data at the customer interface. Personalization and addressing of individual customer needs, for example in the form of next-best offers based on comparable customer profiles, could be even more effective. Corresponding recommendations are based on Big Data in conjunction with AI methods that take into account, among other things, previous purchases and the purchasing behavior of comparable customer groups.

In particular, if the solvency of the customer is known and higher-value products and services can be sold to him as a result, additional earnings can be increased in this way.

Intelligent document processing

The manual, paper-based processing of documents at banks, e.g. financial and accounting documents, requires a great deal of time and personnel. Since numerous tasks, especially in back office, are regularly repeated, are based on fixed rules and the underlying data can be easily digitized and structured, many processes can be automated by artificial neural networks (KNN), such as the capture and processing of documents, the transfer of knowledge during account assignment to new posting records and the reconciliation of account movements with incoming and outgoing invoices. Natural Language Processing (NLP) technologies, for example, are suitable for capturing invoice information.

Intelligent Robotic Process Automation (IRPA)

Robots already determine entire production lines in the automotive industry: The machines work efficiently, with consistent quality and can be used flexibly depending on the technical specifications. These characteristics can be transferred to the service or administration area by Robotic Process Automation (RPA). The application systems are operated by so-called software robots instead of by administrators.

Predestined are application cases that are frequently repeated, occur in large numbers, are controlled by laws, regulations or clearly defined business processes and contain only a few exceptions that have to be handled by humans. The software robot behaves like a clerical worker. The system uses, for example, a virtual keyboard or a virtual mouse. The software robot docks on to the user interfaces and surfaces of the systems and executes the work steps in the same way as the human administrator has previously carried them out. An essential feature of the technical solution is that the application systems used so far remain largely untouched. Only the operation, which was previously carried out by clerks, is replaced by software.

The use of AI extends the field of application of Bots. Intelligent Robotic Process Automation (IRPA) includes applications that understand natural languages, can interpret structured and unstructured data and have cognitive learning capabilities. One example is the automated processing of customer dialogs to arrange service appointments or identify customer requests in chatbots. For example, Deutsche Telekom successfully uses Bots for customer services on a large scale [Abolhassan, F. (2017)].

The IRPA concept also makes it possible to use several software robots in one application area. For this purpose, a robot controller assigns the individual processing cases to further software robots. It analyses the cases according to content criteria, e.g. in the case of incoming e-mails for indications of complaints, inquiries, consultation appointments or usage aids, and delivers the e-mails to the responsible robots for processing.

IRPA projects have a high benefit for process automation in the area of front and back office at banks and should therefore be an integral part of the digitisation strategy. What is unacceptable here is that there will always be special cases that have to be controlled and processed by people in manual processes.

Bottom line

The financial sector is still under considerable pressure to change. Decreasing margins not only require the reduction of costs and the efficient management of risks, but also force institutions to develop new sources of income. AI methods and techniques shape new products, services and business models. The targeted use of AIs along the entire value chain creates potential for increasing revenues and productivity. In addition to the advantages of AI technology mentioned above, risks must also be taken into account. One risk, for example, is that self-learning algorithms cannot be audited and that the industry is increasingly dependent on a few AI specialists and technology providers operating outside the regulatory framework. The growing interconnectivity between financial markets and banks can also lead to systemic risks: Algorithms, for example, manipulated the volatility index at the beginning of February, causing the New York Stock Exchange to plummet.

The prerequisite for banks to be able to realize added value from AI systems is good data quality. Only a high data quality enables the intelligent systems to derive correct forecasts and recommendations for action.


The author Prof. Dr. Dirk Neuhaus, MBA is Professor of Information Systems in Financial Services.


Prof. Dr. Dirk Neuhaus, MBA
Professor of Information Systems in Financial Services
Hochschule der Sparkassen-Finanzgruppe
Simrockstr. 4
53113 Bonn

Phone: (0228) 204 – 9936
Fax: (0228) 204 – 9939

Related posts