Culture, geography, politics, and media reporting all create challenges for name matching in a KYC or due diligence process, even when software is used to perform the matches.
According to Moody’s, a fundamental theme underpins know your customer (KYC) risk screening: names. A person’s name is their key identifier and in screening processes names are of course vital, but names are very rarely unique – unlike the people who claim them.
In the US, for example, topping the chart for most popular first and last name combination is James Smith. And more than 70m people share the surname Devi in India, the most populous country in the world.
In relation to adverse media screening, which is used to find risk associated with negative news stories pertaining to an individual, the driving force behind generating accurate alerts is matching based on a name. As this process can generate many results, Moody’s KYC solutions consider several identifying components, including things like an individual’s location and date of birth, to refine results.
Names can be complicated. Name matching is especially difficult in a global dataset of names captured by sources including government watchlists and adverse media from different countries — including names in more than 70 languages and numerous scripts. The complications grow when the names include those of risk-relevant persons and organizations, some of whom may be deliberately trying to avoid detection.
To manage this process, Moody’s has established an intelligent screening solution supported by a team of cross-functional screening specialists. The team combines qualitative anthro-linguistic expertise with access to tailored quantitative data science algorithms and machine learning techniques to overcome name-matching challenges and deliver accurate results.
Some cultures maintain significant or symbolic names for generations, leading to names that are common to vast numbers of people. For example, middle names in Vietnamese cultures tend to correspond with birth order, offering less identifying information than an additional name component generally would.
Such customs raise questions around how best to capitalise on distinct elements of names in name-matching software design to avoid overwhelming screening results with false positives.
Moody’s has three main avenues for identity to dismiss ambiguity among common names:
- Grid profiles maintain high-quality standards on identifying data, including date of birth, addresses, and aliases, particularly among premium content profiles enhanced with entity-based research.
- Customisable filters for alias, address and date of birth analyse and rule out matches with conflicting identity information.
- Review provides additional post-match research on Grid profiles to reduce alerts to high-confidence matches.
Recording a name from one medium to another, from paper to computer for instance, can easily introduce errors — especially when the name is transliterated. Who audits how an entity name went from point A to point B? Audio-to-text transcription adds many problems but relying on a signature to spell a name can also be problematic.
Luckily, with modern word processing and AI technology, we don’t often have to rely on handwriting to spell a name. However, the adoption of digital records, literacy rates, and access to computers remain uneven between and even within countries.
Name recording may rely on optical character recognition (OCR), which gives near-accurate, yet sometimes imperfect interpretations of written documents and letters. Even where digitisation is widespread, software can introduce errors — for example, automatically ‘correcting’ names – ‘Kara’ to ‘Lara’ or ‘Park’ to ‘Mark’ or ‘Teh’ to ‘The’.
Identifying a misspelling across languages and scripts makes disambiguation less intuitive than it would appear to be. The range of keyboard templates adds to unpredictability. Languages that use the same alphabet may use unique keyboard layouts, like French and English, whereas logographic scripts might be subject to the ‘Wubi Effect’, where several competing keyboard configurations are in common use among a community with a single common language.
Languages, scripts, writing standards, and cultural pressures differ between regions and change over time. Political power plays a surprising role in how names are written, recorded, and transliterated. For example, imposing the Russian language and Cyrillic alphabet in the Soviet republics, as a tool for social cohesion.
In other cases, it’s about cultural revitalisation. The once near-extinct Gaelic language was revived after Ireland became independent in 1922. As recently as 2005, a law came into force that mandates Gaelic versions of place names in parts of the country’s west on road signs and official maps.
What makes name matching even more complicated is when the names of one person don’t match — and by all linguistic means should not match. Some people may not want to be identified, but name changes also have other justifications. Legal name changes are more common in some places than others and are often subject to regulation.
In 2005, South Korea’s supreme court allowed name changes under certain conditions, such as if the name resembles an offensive word or is the name of a famous criminal. Under this law, a person cannot change their name if they have a criminal record or history of bankruptcy. Elsewhere, an individual might go by a different name due to cultural pressure or a desire to assimilate, or perhaps to make it easier for others to pronounce their name.
Moody’s Grid profiles capture all available aliases as recorded in adverse media coverage to ensure identification of risk entities. Further research is included with our high-risk premium content sets, such as politically exposed persons (PEPs) + family or close associates. Review screening also includes critical watchlist logic and optional OFAC-search style algorithms to detect high-risk sanctioned entities.
5 steps to overcome name matching challenges using intelligent screening
Going from many names to a few and then arriving at a final alert can be done with intelligent entity screening processes in the following five steps:
- Conduct initial searches: Use multiple search algorithms to generate a broad list of potential matches based on the names you are screening.
- Rescore results: Refine the search results by considering the cultural context of names, applying specific rules for certain cases, and making necessary adjustments based on additional information like date of birth (DOB) or addresses.
- Apply filtering: Narrow down your results further by implementing firm-specific rules and apply custom filters that account for the severity and types of risks associated with the entities.
- Review: Use our AI system to score the alerts and automatically suppress false positives. Review the remaining matches against custom guidelines specific to your firm.
- Make alert decisions: Have analysts review the shortlist of potential matches. They will make final decisions based on the integrated system’s data and the context of the inquiry.
This process is automated, leverages our screening database with more than 21m profiles, harnesses machine learning, and keeps humans in the loop for decision-making. We have designed the process with the aim of reducing the number of false positives generated during screening and ensuring the alerts raised for further review are those most likely to be true matches.
Copyright © 2024 RegTech Analyst
Copyright © 2018 RegTech Analyst
