We hosted a discussion between Jeanne Le Garrec, Regional Director, ASEAN region at Basis Technology and Hwee Kun Ho, Head of Sales, APAC at Napier to discuss challenges in screening for financial crime compliance and share Jeanne’s vast experience.
In this episode of FinCrime in Focus: Challenges in Screening, we focused on screening challenges: their causes, and how to overcome these with smart technology and a holistic, risk-based approach.
Watch the video here, or keep reading for a summary of the key points discussed:
False alerts are a major screening challenge for financial institutions
False alerts are a real issue during screening at customer onboarding and for transaction monitoring. While the task of verifying if a name is on a watch list sounds very straightforward, false positives and false negatives can quickly make this task burdensome and time-consuming, especially if they are occurring in large volumes and overwhelming the analyst.
"Name matching is a key anti-money laundering [AML] process, but it also creates one of the biggest challenges for compliance officers. The ability to reduce false positives and negatives requires not only a good collection of customer data, but also a high-quality risk intelligence database coupled with a strong name matching engine." - Hwee Kun Ho, Head of Sales, APAC at Napier
What are the root causes of false positives and false negatives?
1. Rigid name matching systems
These usually rely on one method which is rules-based and includes a translation step, which is not recommended. The volume and variety of names adds complexity since there is no universal way to write the same name. For example, some names can be written in both Arabic and Latin script. Humans make errors, such as typos, and machines can modify names. Very few systems can match names from different scripts, so when the name variation is different to what your system knows this creates a false negative, which is not only dangerous but expensive to investigate and ultimately discount.
2. Legacy systems which are not intelligent
These systems return all the results that look similar, from close or from afar, to the name being searched, which creates false positives. Many banks deal with more than 90% false positive rates, so it’s a huge burden on compliance teams and resources. In these cases, the system isn’t able to triage and disqualify false positives without analyst investigation, leading to the scenario of looking for a needle in a huge, time-consuming haystack.
3. Generic systems
These systems are often trained on generic datasets with the aim of being satisfactory for multiple languages, however there is no linguistics differentiation. This is an issue because analysing an English name cannot be done in the same way as a Spanish or Chinese name. It is important to consider the origin of the name/script. Issues also arise if the data is different to the type of data the machine learning model has been trained on. This can lead to extremely poor results and cause false positives and false negatives.
Regulators expect more from screening
False positives and false negatives are the number one challenge for screening and transaction monitoring, which generates excessive alerts and adds to the evolving regulatory requirements.
Many financial institutions still tackle alerts with manual observations. This effort is not only tedious but vulnerable to error. In some cases, the alerts are tackled with tools built in-house, but these cannot keep up with the growing regulatory responsibilities as regulators are increasingly expecting to see more robust transaction monitoring systems.
Improving KYC reviews
KYC reviews are an important but major operating hurdle for most financial institutions because of the amount of information they need to extract and analyse. Taking a risk-based approach to KYC has also become a huge struggle for financial institutions because it has become so difficult to keep up with the frequency of reviews.
KYC processes have traditionally been manual with lots of tedious information gathering needed to conduct risk assessments. The natural choice has therefore been to hire more people, but this is not a long-term solution since it’s inefficient and budgets are always limited. This is why financial institutions are looking for options to enhance and automate the KYC process.
The fundamental motivation for financial institutions is that the cost of compliance is rising and to address this, they need to adopt tech that can help tackle operational inefficiencies and ineffectiveness, while also being agile enough to adapt to new threats.
A holistic risk-based approach bridges the gap between KYC and customer behaviour
Bringing together KYC and behavioural data creates great value in terms of customer insights and analysis, enabling compliance teams to better manage risk appetite and mitigate financial crime risk. This is important not least because regulators expect a more stringent and proactive process for assessing customer risk.
Financial institutions are required to have an end-to-end AML workflow to ensure seamless and comprehensive policy enforcement with informed decisions. All of this can only be achieved with a complete, holistic customer view.
A good screening system makes life easier
A screening system which is adaptable, secure, and user friendly will make life easier. Compliance officers need to justify their choices and results to the regulators, and there’s increasing recognition that AI needs to be explainable.
For compliance teams to have peace of mind, AI also needs to be accurate, tuneable, and explainable. Having a flexible tool that adapts to the uniqueness of the data, the singularity of the process, the bank’s risk appetite, and the evolving compliance rules is paramount.
Smart systems can help overcome false positives and false negatives in name matching
The first way to avoid false negatives is to get a system that can analyse a name in its native form. This can mean applying different AI models based on your data, such as name origin if analysing a person/organisation’s name. The model needs to support many different variations, misspellings, titles and so forth, and should be trained using complex languages like Japanese, Chinese, Korean and Arabic.
To reduce false positives, it really helps to apply the right linguistics models to the right AI models to triage results and bring the numbers down. A weighted, multifield search that considers other data attributes like date of birth and address really helps.
The ability to reduce false positives and false negatives requires good collection of customer data coupled with a high-quality risk intelligence database and a strong name matching engine, all working together. It’s also important to work with best-in-class tech partners and leading data providers like Refinitiv, Dow Jones and Lexis Nexis.
How does Napier meet all the evolving compliance requirements?
Napier is uniquely positioned to transform screening processes and take a more client-centric approach. Its intelligent compliance platform combines new technologies including cloud-based and big data architecture, with AI and machine learning capabilities to efficiently and effectively meet workflow demands while exceeding regulatory expectations.
About Basis Technology
Basis Technology is the leading provider of AI software solutions for extracting meaningful intelligence from multilingual text and verifying identities. Its software Rosette (a Natural Language Processing suite of capabilities) is trusted by the world’s largest and most critical banking and security systems.