Contributed by: Véronique Van Vlasselaer, Bart Baesens
Véronique Van Vlasselaer graduated as doctor in Business Economics at the KU Leuven in 2015 under the supervision of prof. dr. Bart Baesens. Her Ph.D. is oriented towards the development of fraud detection frameworks and solutions from a data science perspective. She enriched traditional fraud detection models with insights achieved from Social Network Analysis (SNA) by featurizing the relationships among fraudsters and non-fraudsters. Together with prof. dr. Bart Baesens and prof. dr. Wouter Verbeke, she is co-author of the book “Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques: A Guide to Data Science for Fraud Detection”.
Since November 2015, Véronique has been working at SAS, the global leader in analytics and artificial intelligence. She supports the Belgian and Luxembourg pre-sales team guided by her team manager Yannic De Bleeckere and country manager Jeroen Vangodtsenhoven.
Can you briefly describe your current job?
As a pre-sales analytical consultant, my main responsibilities are to help companies to get value out of their data by (1) sensitizing, (2) advising, (3) coaching and (4) implementing machine learning (ML) and artificial intelligence (AI) algorithms. I have an extremely varied job: I present for large audiences of business (and/or academic) people evangelizing the data science message and stressing the importance on how to effectively use ML and AI to enrich day-to-day business processes; I give technical master classes for data scientists on how to get started with an analytical project, the challenges and the pitfalls, and how to optimally implement and deploy data science; I regularly visit companies and organizations and organize workshops to help to prioritize and concretize their analytical exercises together with their analysts, end users and business managers, resulting in a roadmap to analytical success, and coach their analysts in the executing of it. But most of my time, I do what I love most: bringing data science to life and literally prove what value data can offer for organizations. During so-called PoCs (Proof of Concepts), a SAS team consisting of a data engineer, a data scientist, an IT gourou and an account manager assists our customers in demonstrating the power of data science and how their business processes can be much more streamlined by integrating analytics.
What types of analytics are you predominantly working on?
Within SAS, I am an all-round analytical consultant, meaning that I can be assigned across industries ranging from banking & insurance, manufacturing, public services, retail, etc. However, my main focus is still on why and how to use ML and AI for fraud and anomaly detection. As fraud is present almost everywhere, my analytical skill set to design such detection systems is challenged every day. For example, fraud in banking is completely different from fraud in public services. Each application domain requires its own tailor-made feature engineering, machine learning algorithms and deployment. While the implementation of machine learning algorithms and the deployment of them are a concern of data scientist and IT, feature engineering is a process that needs the involvement of the business expert. Their input is an indispensable asset in the creation of an effective detection algorithm. Lots of my time is invested in better understanding the business and translating their input in analytics-worthy features. Past projects have shown that only the combination of their expertise and the ability of machines to automatically learn will eventually lead to good fraud and anomaly detection solutions.
What do you consider important challenges for analytics to succeed in the business areas you are working on?
Nowadays, ML and AI is recognized by top management as an important future strategy to support decisions and steer business processes. A recent quote by HBR says that “Over the next decade, AI won’t replace managers, but managers who use AI will replace those who don’t.” I truly support this quote, but at the same time I want to underline that an ML- and AI-driven organization is not something that comes overnight. To be effective, the whole organization has to know what ML and AI can, but also cannot do. Organizational awareness and the development of a trust culture in machines is a very important attention point for management to book success. Building an AI roadmap by prioritizing the business process that might benefit and generate additional €’s from data-driven insights is a first step towards this success. Another aspect that is almost always forgotten, is that ML and AI will not bring any value to the organization until it is deployed, i.e., until it is part of any business process. What I see in many organizations, is that analytical tools are merely considered as nice toys to keep the data scientist busy, but there is a high lack of operationalizing their efforts. I see it as part of my mission to sensitize people that analytics is an iterative process that needs data as input, extracts meaningful insights and put these insights in practice.
Many people are talking about deep learning these days. Do you see that being applied in your field? What is your take on this?
Basically, deep learning algorithms are the Ferrari’s amongst the machine learning algorithms. It boils down to a very fancy and advanced neural network with many hidden layers and lots of connections between the neurons of the neural network. In early days, research has already proven that neural networks are universal approximators, meaning that they can map any input features to the output as long as they are complex enough. Although they are extremely powerful, they come at a cost: the training process of such deep learning algorithms is often very time consuming and might require the use of more expensive GPU’s instead of the traditional CPU’s. The most well-known and successful example of deep learning is image recognition. Although practical applications are still limited, we educate more and more of our customers in the opportunities that deep learning can bring and we see that they start to successfully experiment and discover the possibilities. Deep learning algorithms are without a doubt an important enrichment of the current ML landscape in the years to come… but let’s not overrate it. Deep learning is just a means – like any other ML and AI algorithm – to get to the business objective.
How did a PhD contribute to your current job?
First of all, it needs to be said that doing a Ph.D. in the research group of prof. dr. Bart Baesens is different from a standard Ph.D. As a supervisor, prof. Baesens urges for collaborations between academics and business in the form of research chairs. Such research chair allows a Ph.D. researcher to combine the more theoretical aspect of a Ph.D. with a business part. Insights found during his/her research can be directly investigated and verified by the business partner. During my Ph.D., I had the pleasure to collaborate with the Belgian Social Security Institution. The objective of the Belgian Social Security Institution was to find a streamlined approach to detect fraud networks in the Belgian corporate landscape of companies that are intentionally lacking to pay their taxes, and thus commit so-called social security fraud. From an academic perspective, the focus was on how social network analytics can improve fraud detection algorithms, an approach that was not yet researched and had no proven value in fraud. Because of this collaboration, I did not only learn to work autonomously and with a vision, but also to make it valuable for the business partner, essential characteristics in life. During my Ph.D., I was confronted with many challenges that also frequently occur with whatever analytical project – e.g., data availability, quality, readiness, interpretation, correct understanding of the business input, how to present the results to the business, etc. – giving me a very rich and valuable backpack for my future professional career. It is often said that a Ph.D. is a niche market, drowning into the specifics. However, due to the exquisite guidance of prof. Baesens, I have learned to see the bigger picture: you first need to have a vision, where you want to go, what you want to achieve, and only if you have a solid and well thought trough vision, the specifics will automatically follow. This is exactly how my Ph.D. is built and there is no single day in my professional and personal life that I do not apply this. I am extremely proud of the work that I have done during my Ph.D., but every humble (wo)man will acknowledge that without a great leader s/he would have been nowhere.
