Artificial intelligence in use for compliance

Due diligence requirements are difficult to meet for some banks, in particular because of the vulnerability of commercial transactions to the risks of economic crime, money laundering or terrorist financing. Banks are therefore considered to be the first line of defense when monitoring transactions. But are they equipped to fight fraud and economic crime at all?

According to the Federal Situation Report on white-collar crime, fraud accounts for around half of all cases in the overall white-collar crime sector. Because of their sophistication, modern fraudsters are called "scampreneure" - a neologism of "scam", English for fraud, and entrepreneur. They are constantly looking for new ways to exploit weaknesses in the system, in processes, but also human weaknesses of bank employees. Common types of fraud include, for example, the double financing of invoices, dual use of goods or fraud in factoring. In the latter case, a company issues an invoice to the customer and sells the receivable to a financial service provider at a discount, even if the goods or services have not yet been delivered or rendered to the customer or not completely. This type of refinancing can also be used with fraudulent intent by selling receivables that are not recoverable - this procedure is called "Fresh Air Invoicing". Such fraudulent arrangements between buyers and sellers to the detriment of banks are a common scam.

Artificial intelligence makes know-your-customer processes more reliable and efficient

Exposing the "scampreneurs" and preventing fraud is one of the primary objectives of risk management in banks. The same tools and technologies that banks use to detect money laundering and terrorist financing can be used to do this. Know Your Customer (KYC) processes are at the heart of fraud detection. They require not only transparency and access to essential information, but also effective data processing. Many trade financing transactions do not come about because due diligence - especially in connection with KYC - is not sufficiently fulfilled. This is due not only to the increasing flood of data, but also to the tight resources in the banks.

Artificial intelligence (AI) can help compliance teams to better evaluate the data and identify customers and their activities. But the situation of customers can change over time. Depending on the market situation, companies change their strategy and switch from one-off sales to subscription models, as is currently the case with many software providers. Banks must then be able to capture these changes to continue to detect irregularities without increasing the number of false alarms. Many banks see the greatest potential of AI for KYC in better detecting suspicious cases and reducing false alarms. After all, these are currently one of the biggest challenges facing compliance teams in banks.

AI can detect anomalies more reliably than rule-based systems. An AI-based solution can analyze every unit involved in a transaction. This ensures, for example, that the customer really wants to make the payment and the payee is legitimate. To do this, it identifies data points within a transaction. It also learns which activities are normal on the basis of previous transaction patterns, observes customer behavior and detects any deviations. On this basis, the AI software determines which deviations are uncritical and which require more detailed investigation. In the event of such an alarm, the solution can provide relevant contextual information that enables responsible employees to assess the case.

AI can also help update and classify a customer's risk profile to ensure ongoing compliance. AI technologies can identify large amounts of data - including unstructured data such as text and images - and use machine learning to understand their meaning. As a result, such solutions can very quickly create comprehensive, accurate and verifiable risk profiles of companies and individuals. This can be extremely useful for compliance departments that need to understand complex data about shareholders, beneficial owners, managers and employees. They can use AI-based risk profiles to make better decisions for a risk-based approach to compliance. With more and more international transactions and high-profile cases like the Panama Papers, the importance of AI for identifying the beneficial owners of companies for bank compliance continues to grow. In the case of the Panama Papers, eDiscovery software was used, using Continuous Active Learning - a form of machine learning. This can learn which documents in large, unstructured data sets are relevant to a specific question. The software is taught by trained reviewers and can find further relevant documents on this basis.

Know Your Transaction elaborates KYC

KYC helps to comply with due diligence obligations on a risk-based basis and to take a closer look at more critical customers and their payments. However, KYC alone is not enough. While the quality of customer data has improved over the years, transaction data is still very poor. The information on participants contained in payments is often very sparse and little standardised. With SWIFT MX and ISO 20022, improvements in structure and format are on the way. But the contents of the transaction data are often hardly suitable for drawing clear conclusions about what the payment is really about. The real purpose and origin of payments are often difficult to identify. To identify potentially risky transactions, an automated enrichment of transactions with accurate and relevant information directly from the original data sources would be necessary. Examples of such data are contract documents, customs documents or invoices, if they are purchases. But IP addresses in online banking or the use of GeoData can also provide information about the origin of the payment.

In the area of trade finance, it is also possible to check the consistency of documents for trade transactions with certain document packages, e.g. trade contracts, inspection or packing lists. AI can now help determine whether the transaction involves risk. For example, by comparing it with available market data, AI can identify an unusual price for a particular type of goods.

AI decisions must be comprehensible

A critical aspect of using AI for KYC and KYT is transparency. Some technologies such as neural networks or deep learning sometimes produce results and conclusions that are no longer comprehensible to humans. However, in the area of fraud detection, which can also have legal consequences, this should always be the case. AI methods such as linear regression or Bayesian learning, on the other hand, are transparent and their logic can be understood by humans. By disclosing the most important underlying data elements that control the model, responsible employees can understand the reasons for the assessments made.

Outlook

Artificial intelligence can support compliance teams in banks in many ways. The technological possibilities already exist and are developing rapidly. However, the legal framework for their use must also be further developed. After all, automation with the help of AI raises similar questions in the banking world as it does with autonomous vehicles: What decision should be taken in a dilemma case? How autonomously should the technology act? And who bears the responsibility?