The Centre of Excellence for Data Sharing and Cloud (CoE-DSC) has completed a case study on the application of Artificial Intelligence (AI) Entity Resolution technology to link records across datasets without exposing the source data. In this case study, our participants Knights Analytics together with Roseman Labs tested the performance of their AI Entity Resolution technology combined with Multi-Party-Computation (MPC) – to improve financial crime detection by enabling banks to generate combined views on transaction patterns and the entities involved. The CoE-DSC presents the report with metrics showing that MPC combined with Entity Resolution limits data exposure while achieving solid performance in comparison with today’s practices. Lastly, the report considers potential implementations of this technology for other sectoral use cases.
AI Entity Resolution has potential to support the combat against financial crime by linking identities across datasets.
Financial crime is a severe problem and combatting it is essential to protect society against harmful effects. (To illustrate, an estimated €2.2 trillion proceeds from activities such as forced prostitution, terrorism, and drug trafficking are laundered yearly worldwide1). Tracking financial crime requires banks to collaborate2 to gain a combined view on actual transaction patterns and involved organisations. To create such a view, records related to organisations need to be linked across datasets from multiple banks. The procedure of matching records to an organisation or a person is called Entity Resolution. It helps to uncover the real-world identity behind the records. For example, using Entity Resolution one can connect affiliations, bank accounts, or beneficial owners to a customer record. With the help of AI, banks can conduct Entity Resolution in an automated, and efficient way when dealing with millions of records, disparate attributes, and absent unique identifiers for matching.
Banks face a dilemma on how to ensure privacy of sensitive data to gain enhanced insights by linking datasets.
Conducting Entity Resolution across banks is hard, since data with sensitive and confidential information cannot be shared directly due to privacy regulations, confidentiality agreements and professional secrecy obligations3. Thus, this analysis needs to be conducted using technologies that minimise the usage of sensitive data while producing meaningful insights in an efficient way.
Entity Resolution with MPC generates insights on transaction patterns of organisations involved without direct data sharing between banks.
Multi-Party Computation (MPC) is an emerging technology in the domain of data sharing, which allows multiple organisations to collaborate on data and gain insights, while keeping all input data private4. Implementing Entity Resolution with MPC can help with combatting financial crime because it enables banks to generate a detailed combined view, without risking exposing privacy-sensitive data5.
Tests show that when combining the Knights Analytics Entity Resolution algorithms with MPC, the Entity Resolution performs well in terms of both quality and scalability.
In this case study the performance assessment of Entity Resolution with MPC was conducted by Knights Analytics on real company data provided by OpenCorporates. In these tests, AI Entity Resolution was first applied in the clear, which served as a baseline for comparing how well the solution performs when combined with MPC. The results show that MPC vastly reduces data exposure, while performing nearly as good as when running the solution without MPC in terms of quality, while also overcoming key scalability challenges in MPC. To achieve this result, data needs to be uniformly pre-processed at each bank, because performing Privacy-Preserving Entity Resolution on the raw data without pre-resolution won’t yield satisfactory performance. See the results’ overview below, more details can be found in the report.
Table 1. Summary of the performance results
Other sectoral use cases can benefit from AI Entity Resolution technology with MPC
Based on this case study, the CoE-DSC developed an assessment framework to discover relevant use cases. For example, the tested technology works well for use cases where obscure records with no unique identifiers need to be linked, and sensitive data is fragmented over multiple parties, Please, read below use cases for child protection, prevention of tax evasion, and detection of insurance fraud, also included in the report.
Table 2. Other sectoral use cases
Findings of this use case
Do you want to learn more about the use case design, insights on the context of the use case and insights on how the use case can be implemented in a scalable way to other use cases with similar roles? Please download our report which includes our most important findings so far.
Do you want to know more about this use case? Or do you have an interesting idea to define and realise new cross-sectoral use cases of data sharing? Please send us an email at email@example.com.
References: 1 World Economic Forum on financial crime 2 European Court of Auditors Report (2021) 3 EBA Confidentiality and Professional Secrecy Assessment Principles (2022) 4 UN Guide on Privacy Enhancing Technologies – Section 2.1. 5 Het Financieele Dagblad on technologies helping collaborate to combat the financial crime (2023)