Compliance, internal control and risk management functions have become very consuming in terms of human resources. The percentage of a financial institution staff dedicated to these functions varies from approximatively 2 to 6, according to the nature of the activities and the risks incurred.
Thus, the problem of the task automation of these monitoring functions arises and brings up the topic of the artificial intelligence contribution to these professions. If we consider the fight against money laundering for instance, it is now obvious that the staff numbers are constantly increasing but yet incapable of meeting the challenge of fulfilling all the tasks assigned, in particular concerning the analyses of suspicious transactions which themselves originate from tools set up with different criteria supposed to avoid manual analyses of all the customers’ operations.
Nonetheless, the number of these alerts is often very large and requires significant resources assigned to the verification of the alerts before reporting them to TRACFIN or dismissing them due to an insufficient collection of justifications.
This analysis remains largely manual when it could be mostly automated. This analysis indeed goes through standardized, even “routine”, stages: analyzing the coherence of the given justification for an atypical operation or requesting adequate documenting evidence if need be.
In a report of Cédric VILLANI entitled For a meaningful Artificial Intelligence, four elements are mentioned as bottlenecks of the automation:
When we analyze the future automatization possibilities of compliance and monitoring activities, we must take into account the routine analyses which can easily be automated, the possible obstacles to a total automation and this analysis will also allow an identification of possible impacts on employment.
Today, monitoring is still incomplete due to a lack of staff and some money laundering alerts often take too long before being treated.
In some financial activities, industrial processing of files (consumer credit for instance), the number of checks needed to obtain a representative sample also implies an automation if we want to reach the result expected.
The artificial intelligence thus appears a way to go further in terms of control coverage without necessarily destroying existing jobs.
Moreover, a compliance officer whose analysis work is partly undertaken by a machine will surely spend more time on a finer analysis and exercise their “cross-cutting cognitive skills” mentioned above.
The issue of using artificial intelligence in the risk, monitoring and compliance sectors is therefore particularly significant. Nonetheless, we will have to make sure that these algorithms do not become black boxes. Mastering and auditing those algorithms is a major issue for these control functions.
President of Regulation Partners.