‘Former police chief of Houston once said to me: “Frank Abagnale could write a check on toilet paper, drawn on the Confederate States Treasury, sign it ‘U.R. Hooked’ and cash it at any bank in town, using a Hong Kong driver’s license for identification”’ (Frank W. Abagnale, Catch Me If You Can: The True Story of a Real Fake).
Frank Abagnale apparently had no problems forging checks in the 1960s. The world has moved on significantly since the ancient Greeks and has become much faster, more networked and also more data-rich. The power and complexity of stored customer and transaction data makes it increasingly difficult to detect fraudulent acts. For individual institutions it is extensive networks in particular that often elude identification.
Nevertheless, there are tools that can help them detect fraudulent acts as reliably as possible (see graphic): forensic audits, human intelligence, digital forensics, research on open source platforms and data analysis.
In the following, we will take a more in-depth look at data analysis as a potential fraud detection tool.
Data analysis is a wide-ranging field and its outcomes are heavily dependent on the quantity and quality of the raw data. The purpose (real-time/trend/network) is also crucial. The following outlines some of the simpler methods – plausibility rules and statistical methods – that are already being used in many enterprises, together with new technologies such as data mining and network analysis.
Plausibility rules: As a general principle, plausibility rules can be implemented for several systems within an enterprise, either for real-time detection of suspicious transactions or for identifying general discrepancies in customer master data. Viewed in isolation, these rules do not provide a comprehensive insight into the data of individual customers or customer groups.
Statistical methods: This approach examines a range of individual aspects for either customers or customer groups as a whole. For example, regression or distribution analysis can be applied to all customer master records to identify anomalies.
Data mining: Data mining refers to intensive data analysis incorporating aspects of artificial intelligence for detecting hidden patterns. This type of analysis requires large volumes of data. The data is first examined for irregularities (missing data or data errors) and then subjected to association analysis (for example, someone purchasing a toothbrush may also buy toothpaste). The available data sets are subsequently classified by clusters. Constantly recurring data sets are applied to the generally applicable structures. A statistical regression function with MSE reveals the relationships between data and data sets. Data mining is used to identify patterns in customer behaviour and, where appropriate, to expose them as fraudulent or at least suspicious acts.
Network analysis: Network analysis is used to identify hidden networks – i.e. a person with different identities or criminal networks consisting of multiple individuals. It is implemented using graph algorithms. Correctly mapping the network requires a definition of the ‘nodes’ and ‘edges’. Here nodes may represent a wealth of data sets, for example name/mobile phone number/smartphone manufacturer/GPS data, first name/last name/address or first name/last name/account number. Using fuzzy logic and data clustering, it is subsequently possible to create connections between ‘individuals’ who appear not to be related at first glance. For example, this could be one and the same person who uses various spellings of their name (ph = f) and who is registered at different addresses. Further networks are created by links between individuals (for example because of account transactions or friendships in social media, etc.).
Network boundaries and the identification of network structures constitute the major challenges here, requiring comprehensive, overarching data volumes from different sectors (public authorities, telecommunications, social media, banks). At present, network analysis can only be performed at the official level, which allows comprehensive data insight.
Thanks to state-of-the-art technology, we can catch Dolos before Aletheia’s twin – the personification of lies – is placed in the kiln and brought to life. And we can prevent the twin from stealing part 3 again.