Niels Holtrop explains how he developed a statistical model allowing companies to collect consumer data safely for his doctorate at the University of Groningen's Faculty of Economics and Business.

Every time we buy something online, use an internet browser or take out a subscription, companies are collecting valuable information about our preferences and our lives.

This data is used for marketing purposes, for example to target advertising to a specific group. You might have noticed that if you look up a certain product in a search engine, you will soon find advertisements for the same product featuring on the websites you visit.

Each piece of data collected is small, but added together they form a detailed picture of our lives. Governments and consumer groups are increasingly concerned that the collection of this data and the way that it is used could be a threat to consumer privacy. As a response, some companies take steps to severely limit the data they collect, thereby hampering their ability to pursue marketing opportunities

I have recently completed a doctorate at the University of Groningen’s Faculty of Economics and Business. As part of my PhD research, I developed a new statistical model that can balance customer privacy with the marketing opportunities that companies are interested in.

Our model allows companies to collect customer data, but only requires them to store the relevant information (i.e. the parameters of the model) rather than the underlying data. This means they can use this information for marketing purposes, without needing the underlying data.

We tested this model in a typical setting: Predicting the probability that a customer will end a contract after its final date, based on the characteristics of this customer. Whereas before companies needed a long history of information on a customer to do this, our approach stores the individual customer level information for each year in the model parameters, thereby capturing the relevant information in the model itself. This makes it possible to delete the customer data after it has been processed into the model.

This means there is no need for masses of information about clients to be stored in a database for a prolonged period of time. This has proven a serious security vulnerability in the past, as for example hackers have targeted these troves of data. Security breaches have allowed enormous leaks of personal details about millions of customers. For example, in a 2014 breach, hackers stole the data of 145 million eBay customers. Given that data leaks occur more and more (almost 5500 in the Netherlands in 2016), taking this threat seriously is important for companies and customers alike.

Our research shows that our model improves customer privacy, without damaging the company’s ability to market to their customers. We tested our approach with real data from a healthcare insurer, and found that with our model, predicting which customers will leave can be done just as effectively as approaches which rely on long customer histories. It means companies can make sound marketing decisions while limiting privacy violations for their customers.

For more information, see my PhD thesis and the paper No future without the past? Predicting churn in the face of customer privacy in the International Journal of Marketing Research.