Innovation Insider, let's talk with Anne Briec about fighting fraud and protect our clients

I'm delighted to have interviewed one of our top talent in IT, Anne Briec for this interview. Anne heads up the anti-fraud solutions team within the French Networks Security Department and is one of the most experienced professionals within the Group leading the development of a relevant and efficient AI based solution to fight fraud and protect our clients. During the last four years, she has led the scale-up of the well-known MOSAIC offer along with fraud experts, Data Scientists and Data Engineers developing innovative solutions and thwarting increasingly more sophisticated and complex attempts.

Anne, we often talk about MOSAIC as one of the most mature and valuable Data/AI use cases within Societe Generale, could you tell us more about its purpose?

MOSAIC which stands for More Security with Artificial Intelligence is several in-house applications helping detect and prevent external fraud. 

To understand the background of such development, it is important to note that the scale, diversity and complexity of fraud schemes are constantly evolving making fraud more pervasive and increasing the costs on clients and financial institutions. In France only, estimates of fraud’s total amount for payments means has over-reached 1 billion euros during 2018[1]. Several reasons supporting this trend: the shift to digital and mobile customer platforms, the need for quicker or instant execution of payments, the development of open banking, the growing sophistication of fraud techniques (exchange of stolen data, fake websites, dark web...)

As a trusted third-party, the primarily role of the bank is to protect our clients, reduce the financial costs and prevent reputational risk. This is the exact purpose of MOSAIC, we harness the power of big data and Artificial Intelligence to protect our clients by helping detect and prevent fraud on millions of daily banking transactions. We rely on a twofold approach combining an automatic blocking/timing system for some operations and a robust alerting system issuing alerts to fraud analysts with useful insights to better deal with each suspicious case. 

MOSAIC is an interesting illustration of a successful scale-up approach on one single Data/AI use case, could you give us more details on the achievements to date?

It is true that when we speak about MOSAIC today, we designate several applications on a multi-market and multi-brand basis. It is used by four fraud analysis units within the Group. This is the result of a progressive scale-up approach in line with the Group Innovation and digital strategy and supported by an API and microservices architecture which allow agility and flexibility when expanding the solution to new markets and new brands.

When the use case was initiated in 2015, it started as a POC[2] on a small scope within the French retail bank Société Générale before going to Production in December 2016. The solution has then been progressively industrialised and its scope multiplied eight-fold in five years covering different application domains (instant payment, Paylib offer, APIs DSP2, mass payments, etc.), various clients segments (individuals, professionals, businesses and corporates) and brands with “Societe Generale” and “Crédit du Nord” both retail banking networks covered.

 What factors are key to this success and what are your challenges?

For me, this crescendo growth process was mainly based on three assets:

• A sufficient mastery of the data that we exploit: we use data management tools made available at Group level (application bricks, big data technologies, warehouse, APIs, etc.) since the quality of the data and their storage are prerequisites for the proper functioning of MOSAIC applications.
• A solid collaboration with the business teams which set MOSAIC as a requested solution for any new project dealing with external fraud, deemed essential for its start. This was the case of the opening of the instant payment transfer bank service or the detection of external fraud via open banking channels as part of the second Payment Services Directive DSP2.
• The continuous development of big data and Machine Learning tools to achieve our goals with 18 ML models currently integrated into the application.

There are three principal challenges we need to address:

• Detection of new fraud scenarios and the new capacities of fraudsters
• Exploitation of multiple raw data sources
• High availability of the solution (24h/24 and 7days/7)

Where do you see our AI transformation in 5 years?

Betting on digital transformation by investing in Artificial Intelligence is, out of doubt, one of the most powerful tools for companies to gain a competitive edge. However, the use of AI has introduced a set of challenges and made us question a lot of its aspects such as security, data protection, automatization, interpretability of ML models, AI & ethics etc.

Therefore, I believe that in the coming five years, AI transformation should be driven by those new challenges to build more robust AI that can be totally trusted.

3 key words to describe yourself?

Team development, enthusiasm, adaptability.

1: According to the observatory for the security of payments means provided by the bank of France - 2: Proof of Concept

Do you like this article? share and/or comment. And do not hesitate to suscribe to the Innovation Insider newsletter. Let's share trends and insights from leaders, partners and co-workers inventing the future of banking. Not to be missed!