Why Machine Learning Brings up to 57 Percent Savings in the Kyc Process

In the future, KYC processes will be increasingly influenced by technological and quantitative criteria. What effects does this have on the KYC process? We have taken a closer look at the sanctions lists and PEP screening.

Over the past two decades, the compliance function has gained in importance for banks. This is partially against a backdrop of numerous money-laundering scandals and embargo/sanction regimes. In an attempt to prevent money laundering and the financing of terrorism, various quantitative methods have been established, especially the risk-based approach that was recommended by the Financial Action Task Force (FATF).

For this risk-based approach, the use of advanced methods and analysis technologies such as machine learning are on the rise. Here, we delve into the most suited procedures for the analysis of customer data and what benefits can banks and insurance companies derive from these.

Which machine learning methods are suited to improve the KYC process?

There are several machine learning procedures that are appropriate for compliance. Supervised Learning with Random Forests has proven to be a method that is easy to use and understand. At the same time, it offers high quality results.

Banks and insurance companies are the winners: Over 50 percent savings in clarifying false positives in the KYC process

ACTICO Machine Learning allows the Compliance officers to classify the test results in individual cases, so that the random-forest algorithm is automatically optimized. With their feedback, they improve the future classification. The best results are therefore, achieved when compliance and machine learning experts cooperate and train models. As a result, compliance departments save up to 57 percent of clarifications.

How does Machine Learning interact with the KYC software?

Every bank and insurance company uses KYC software to check new and existing customers. This software matches customer data with entries in the sanctions list, the so-called ‘true positives’. But, the exercise also generates false positives, i.e. hits that do not match but are erroneously shown by the system. From a regulatory perspective, this is correct. However, compliance staff must invest a lot of work in clarifying these false positives, even though they are not posing any risk for the business. The fewer false positives there are, the less time and effort compliance staff have to spend on these clarifications. This is where machine learning comes in.

What do banks and insurance companies say about the savings potential?

Data analyses have shown that the cut-off ratio is up to 43 percent. This means that 57 percent of the hits found are highly unlikely to represent a risk and therefore do not need to be clarified. Machine Learning is included as a component in the ACTICO Compliance Suite and predicts the probability with which a person represents a risk.

Conclusion

Machine Learning achieves a prioritization of hits, i.e. matches found during the name check. The identification of false positives and the exclusion of these non-relevant hits considerably reduces the follow-up effort and relieves compliance staff.