Inspired AI Event - Recap

Challenging the status quo

"I will be speaking for the next 10 minutes. During this time, 15k suspicious transaction alerts will trigger and 99% of them will be false positives." As an introduction to transaction monitoring in the context of anti-money laundering (AML) measures, Laura revealed some shocking figures on how outdated current AML processes and systems are. Compliance and risk teams are trapped in monotonous and often pointless work instead of focusing on detecting real money laundering cases and criminal behaviour.

In somewhat simpler times, when banking was done in person, bankers knew their customers much better and were able to assess the risk and opportunities. In this more digital and complex financial world, understanding risk and opportunity is much more challenging. We are now at a time where the same technology which disrupted the industry could now be leveraged to bring customers much closer to financial institutions.

Unlocking data silos

Laura explained the Sentinels approach to processing client data for understanding their behaviour better. The Sentinels approach involves four steps:

• Understanding data sources and unlocking silos.
• Cleaning and structuring data.
• Building and training ML models.
• Translating the results into client-centric insights.

From transaction monitoring to fraud prevention, from credit assessment to marketing intelligence, financial institutions can now finally leverage from their data to not only assess the risks but achieve new levels of growth. Sentinels’ data unlocks the silos, an AI-powered engine helps to make sense of the data and entity profiles allow the customer-centric approach to display the results.

Slimmer AI’s rich history

Sentinels is the financial risk and compliance business of Slimmer AI. The company consists of +60 engineers with more than a half specialising in AI technologies, such as machine learning (ML) and natural language processing (NLP). Slimmer AI has been around for more than 10 years solving more than 100 real-world problems and proving the impact of applied AI. It’s through this extensive expertise that we have finetuned our “Lab-to-impact approach” to AI product development.

Following this brief introduction to our history, Laura went much deeper into the technology behind Sentinels.

Data science and AI techniques in AML

1. Network Analysis

"It is quite fascinating, how much you can learn from your customers by looking at their relations."

The Sentinels Network Analysis functionality helps our partners discover important insights they would otherwise miss:

• New patterns of behaviour.
• Uncovered anomalies.
• Relationships between accounts.

Analyzing one of the customer cases in more detail, Laura showed an example of a transaction coming from a high-risk country which was originally flagged as suspicious, but after a thorough investigation was revealed to be perfectly legitimate. In AML, this is known as a “false positive”. Along with the business transactions, senders and receivers were also flagged and reported as suspicious.

In a second example, Laura showed a lot of triggered alerts, but based on the traditional rules, did not at first glance appear to be a problem. The senders and receivers of these triggered alerts were all car dealers (B2B transactions). Repeatedly, the alerts kept coming back as false positives. According to all of the documentation provided, these transactions appeared to be legitimate, and numbers were in line with what you would expect for these types of business. However, with the insights and connections gleaned from the Network Analysis, the compliance specialists were able to uncover a VAT fraud scheme. Further investigation helped detect that some of the receivers were using the same IP address.

2. Seasonality Correction

"Seasonality is modelled and used to predict the seasonal component of a merchant’s turnover (daily, weekly, monthly). This way Sentinels can disregard the seasonal component as "expected variation" and only trigger an alert if the variation is truly unusual."

Explaining another example, Laura showed how the seasonality correction removed the former positives from the data and helped to decrease the compliance workload by 50%.

Conclusion

Laura wrapped up her presentation on the point where we had started. Truly knowing your customer decreases costs and increases revenue. It not only helps you to prevent activities that might cost you regulatory fines or loses in money laundering but actually show you real commercial opportunities. And the technology has finally caught up to these ambitions, especially with the help of pragmatic AI.

To learn more about Inspired AI go to: inspired-minds.co.uk/inspired-ai/