Canadian Qingchen Wang obtained his PhD from the UvA’s Amsterdam Business School with a paper on machine learning in digital marketing and operational management. Wang investigated, among other things, the data-driven recovery of debt by collection agencies, the optimisation of call centre staffing and the improvement of multichannel conversion among consumers.
Wang defended his PhD thesis, which was completed under the supervision of Prof. Marc Salomon, in June of last year. After 4 years in Amsterdam, he moved to China, where he is now assistant professor of Innovation and Information Management at the University of Hong Kong. Marketing, he states, has in some respects progressed further than other disciplines when it comes to the deployment of machine learning (ML).
At the same time, there is considerable room for improvement. ‘Take, for example, Albert Heijn, which knows a lot about the buying behaviour of its customers but doesn’t know how customers behave outside the shop. For products that haven’t been sold before, you don’t know what interests your customers.’ What he makes clear here is that many individual companies lack important data. Wang is interested in optimising marketing campaigns. ‘Which products, for instance, should you promote and at what prices? Algorithms can support this kind of decision.’
As part of his research, Wang put into practice ML and approximate dynamic programming to help a debt collection agency in Amsterdam to optimise the collection process. He used historical interactions and results to develop a method that could decide intelligently which debtors the agency should call on a given day.
Implementing the method is very easy in theory. But there are also reasons why a company might choose not to apply ML straightaway. ‘The most important is data privacy. It makes sense that people are careful here. It is also the domain of legal practitioners. For research purposes, data privacy is not an immediate issue and privacy fell outside the scope of this research. An organisation like Google might well be quite transparent about how it uses data but, in many traditional businesses, the management and company structure does not yet lend itself to the optimal deployment of data-driven methods.’
A fundamental issue in determining staff levels at call centres is maintaining a certain quality of service (QoS) level at minimal cost. Wang developed a simulation-based ML framework to optimise staffing at call centres. The framework predicts, with a 1-2 per cent margin of error, the expected QoS for even large call centres and generates work rotas for optimal staffing.
Wang observes that, in the Netherlands, businesses are already better than average at solving optimisation problems. He also experienced this when he worked together with a startup that helps car owners in Amsterdam find a free parking space. ‘They work together with building owners that ‘rent out’ unused parking spaces. In Amsterdam, there is more parking space available than people think. This is a typical optimisation problem that Dutch companies are good at. But there is still much to be gained ‒ there’s a good reason why these kinds of problems keep many scientists busy.’
In Hong Kong, Wang is now addressing a completely different challenge: shortening waiting lists in hospitals for people with back problems. ‘In Hong Kong, people have to wait an average of 3 to 4 years before they can get help. But what we see even now is that, initially, many patients don’t get the right help so they often come back quickly. Together with doctors, we are trying to examine why people come back and how we can prevent this with a view to shortening waiting lists. These, too, are optimisation problems where ML can play a role.’
Wang’s interest is broad and he sees challenges for ML everywhere. ‘I sometimes hear about a project somebody is working on. This might be through friends, contacts on social media or colleagues. I’m always very willing to explore the possibilities and to embark on a new project.’