In a paper published in Science, the researchers reveal that the dates and locations of four purchases can be used to identify 90% of people in a data set recording three months of credit-card transactions by 1.1 million users.

The paper's authors note that financial metadata has great potential, helping with credit scoring, fraud detection, and understanding the predictability of shopping patterns. However, Americans are extremely uneasy about how the information is used, with 87% considering credit card data as moderately or extremely private.

To see how identifiable people are by their purchases, the team took a data set of credit card transactions for 1.1 million users in 10,000 shops and stripped out names, account numbers and obvious identifiers.

Despite this, they were almost always able to identify individuals simply by seeing when and where they had, for example, bought coffee. When the approximate price paid of that coffee is added to equation, identification becomes even more likely.

The researchers also coarsened the data, making the time windows for purchases larger, showing geographical areas rather than specific shops, and making the purchase prices vaguer. Although this does make it less easy to identify individuals, it is "not enough to protect the privacy of individuals in financial metadata data sets".

Meanwhile, women are more identifiable than men, and people with higher incomes are also easier to identify, perhaps because of how they share their time between the shops they visit.

Despite the privacy concerns, first author Yves-Alexandre de Montjoye told MIT News: "[Co-author] Sandy and I do really believe that this data has great potential and should be used. We, however, need to be aware and account for the risks of re-identification."