Privacy-Preserving Cross-Bank Financial Crime Amalytics at Scale 

Collaborative cross-bank Financial Crime Analytics brings a promise of higher detection rates with lower false- positives. However, deploying modern Machine Learning algorithms on bulk data from multiple banks runs into complexities of data sharing and exposes long-known limitations of Machine Learning.

After decades of a KYC-centric in-house AML, financial institutions incorrectly assume that cross-bank AML should necessarily involve sharing of personally identifiable information as well. This assumption raises legitimate data protection concerns, which even Privacy Enhancing Technologies (PETs) struggle to address.

Beyond the data sharing challenges, contemporary Machine Learning-based analytics models require accurate labelled data for training and testing, and produce results that cannot always be adequately explained. Given the historical tendency of financial institutions to 'over-report' suspicion, even the most advanced artificial intelligence trained on financial institutions' isolated or pooled data is bound to produce results that reflect the historically poor track record of spotting and reporting complex financial crime patterns.

In this White Paper, we present Amlytic – an alternative, context-centric approach to bulk-scale cross-institutional Financial Crime analytics at the pre-suspicion level. Amlytic emphasises data minimisation on top of anonymisation, limiting the scope of shared data sufficiently to leverage PETs in a privacy-respecting manner, in-line with the GDPR and the expectations of society. Designed to work at bank-scale and viable under the existing legal and data protection frameworks, Amlytic is capable of delivering robust and explainable results by using as little of the underlying shared data as possible, and with no requirement for training data.

Amlytic relies on Privacy-Enhancing Technology provided by Secretarium, as implemented since 2018 as part of the DANIE initiative, a group of banks, data vendors, and other institutions reconciling some of the most sensitive client reference data and securely sharing insights. Secretarium provides the privacy layer that securely manages the obfuscation and integration of each financial institution's minimised data into a linked and de-identified dataset and serves as an interface for post-analysis investigations and collaboration.

Amlytic leverages the Transaction Analytics Technology provided by FutureFlow, as implemented for the UK Tri-Bank Initiative in 2019. FutureFlow runs on the linked and de-identified data provided by Secretarium to automatically spot pockets of suspicious activity, as well as to allow the participating financial institutions to explore their existing cases of suspicion in the broader context of the de-identified cross-bank data.

In 2022, Secretarium and FutureFlow showcased Amlytic as a comprehensive privacy preserving and analytical platform for the ACPR / Banque de France Confidential Data Pooling TechSprint in Paris, where it was chosen as the winning solution, to be deployed by a group of participating financial institutions. In 2023, we redeployed Amlytic for the G-20 TechSprint in Mumbai, where it was chosen as the winning solution for cross-border financial crime prevention.

This paper summarises the factors that set us apart from the competition in these benchmark global industry showcases and demonstrates how a successful country-scale, privacy-forward, and effective AML solution can be implemented today, with production-level technology and under the existing regulatory framework.