In the last couple of years, European banks have received fines or settled charges for record-breaking amounts for insufficient Anti Money Laundering (AML) practices. Scrutiny by regulators has shaken the status quo: all major banks in the Netherlands have indicated that improving Know Your Customer (KYC) and AML practices will have priority the coming years. We provide three different methods on how digital solutions can be leveraged to take KYC and AML practices to the next level.
Natural Language Processing (NLP)
NLP is a technology that enables software to understand human languages. Therefore, NLP combined with AI can optimize KYC/AML by analyzing texts, documents and other public data (for example, public records, news, social media, search engines). It can be used to look for suspicious information regarding the customer, the beneficiary or associated parties. This can be done as part of the onboarding or periodic review, but also continuously.
KYC and AML in short
A KYC process is followed to verify the identity and financial status of potential customers with the purpose of preventing crime (e.g. money laundering or terrorist financing). It includes: checking personal details, financial information and conducting credit checks. AML contains procedures and best practices to prevent and stop the gain of money through criminal activities.
If applied continuously, NLP can have a predictive ability. For example, it can detect when someone transfers money to a person, institution or country that is currently not on official blacklists/sanction lists, yet is already in the news in a negative context. Also: there are so many different ‘John Smiths’ that it would be a difficult and very time-intensive task to manually identify the right John Smith and the corresponding relevant information. A combination of NLP and AI can help to accelerate this process.
NLP understands multiple languages, thus makes it easier for financial institutions to perform more thorough checks and adverse media screening. I think NLP has a great potential to improve KYC, however, it could be that the costs will not weigh up to the benefits.
– NLP supports multiple languages.
– Predictive ability.
Algorithms for AML transaction monitoring
After onboarding, a next step is transaction monitoring. Financial institutions own huge quantities of transaction data that can reveal unusual transactions or indications of criminal activities. Examples are transactions to high risk geographical areas, or a suspicious time, size or volume of transactions. Below, three ways to find transactions that might give indications of money laundering are described:
- Commonly, suspicious transactions are found by manually determined rules (e.g. all transactions of more than 10.000 euro to certain countries are flagged). However, these rules are likely not all-covering. In ever changing financial environments, timely updates are necessary to avoid deterioration of the quality of monitoring.
- A more advanced way to detect irregular transactions are supervised algorithms. They learn and adapt their behavior based on user feedback. If a user flags a transaction as suspicious, the algorithm learns from this and behaves differently the next time.
- The most advanced types of algorithms are unsupervised algorithms. They possess the ability to ‘self-learn’. Unsupervised algorithms can detect unusual patterns in data that manually defined rules would not have found and automatically update themselves when new data comes in.
Algorithms may improve the quality of transaction monitoring significantly, but it is likely that supervisors and regulators will be hesitant in allowing their use, because of a lack of auditability. I think it will take time for machine learning algorithms to be part of everyday transaction monitoring.
– Ability to identify suspicious transactions that manual set rules do not find.
– Unsupervised algorithms update themselves, so no need for constant manual updating
KYC powered by data
NLP and transaction monitoring both contribute to gather data. The more and of higher quality the data is, the easier it will be to perform successful data analytics.
Joint KYC platform
In the future, when joint KYC platforms are established where institutions share client information and documents, more data will become available. Better insights will surface more easily. In addition, best practices regarding KYC and AML can be exchanged. However, I believe that privacy regulations might hinder the formation of such platforms.
Data remediation and standardization
Data from different sources might be used for KYC practices and should be unified in a standardized format. The data quality must be assured and where necessary the data should be remediated (repaired).
Digitization of client documents
Not all documents might be available digitally. Another problem could be a lack of historical data. To increase the completeness of the digital pool of data, paper documents must be digitalized.
– Standardized data formats.
– More available data and higher data quality cause better data analytics possibilities.
Developments regarding data and the use of NLP and machine learning algorithms, can all contribute to more efficient KYC and AML processes. Criminals will always try to find ways to cheat KYC and AML checks. However, by evolving and improving procedures, risks and threats of criminal activities can be reduced