Société Générale Reduces False Alerts by 98%
New communications monitoring system integrates Rosette text analytics to find real business risks in a sea of data
When a bank’s compliance system is returning so many false positive alerts that the true positives — real business risks — are missed, that leaves a financial institution open to hefty regulatory punishment and reputational risk. This was the situation faced by Société Générale, which operates in over 67 countries with around 150,000 employees.
Learn in this video case study how by integrating Rosette text analytics into its communication surveillance compliance platform, Société Générale drastically cut the number of false positive alerts, thus going from from tens of thousands of alerts a day to about 200, from which investigators can find the two or three alerts per week that truly need to be escalated for further investigation.
If we look back 30 or 40 years, electronic communications were relatively rare. In 1979, there were 10 million SWIFT messages. Today on average, we have 32 million SWIFT messages per day in the financial industry. The regulatory penalties for failure to properly monitor employees have risen significantly. Since 2008, banks have paid well over $300 billion in fines for these kinds of failures. So anything that we do not know of becomes a risk for the company.
Like many banks large and small, Société Générale was overwhelmed by the number of false positives that their communications monitoring system was triggering. They were getting tens of thousands of events per day, making it impossible to respond. So they decided to revise and rebuild their system from scratch.
We had to put in place the appropriate controls to track, review, and analyze all regulated communications. We needed to have a holistic view on all regulated employees, both audio and written electronic communications and the associated operations. Furthermore, as we are an international bank, multilingual capability is a requirement.
They knew they needed a more modern system that could monitor a wide variety of information sources and use artificial intelligence and natural language processing to analyze the data.
The platform would ingest all communications from regulated employees. It would then perform analytics against all the communications, in 26 languages currently, and often compare it against operations such as financial transactions. It would also leverage our KYC/AML systems to provide more robust and reliable alerts. On this platform, we have gone from tens of thousands of false alerts to a couple of hundred. Out of those couple of hundred, two or three of them get escalated every week. The end result today is that we are able to provide a social network analysis of a regulated employee whose behavior warrants further investigation, which prior to this was completely impossible.