The challenge

Splitser is an app that allows you to easily offset group expenses. Splitser has a huge amount of user data and wanted to extract more insights from it. For example, they wanted to map what user types use their platform and what their needs are. The ultimate goal is to make the application more user-friendly and find opportunities for new revenue models.
Main challenges here were the quantity and structure of the available data. For example, more than 100 million rows of spending data were available. Moreover, users of the app use different terms for the same expenses; for example, more than 50 different terms or spellings were discovered for expenses related to toilet paper. Who could have predicted that?

What we did

We arrived at a trajectory of two clustering projects:

1. In project 1, the data was explored for finding types of spending (think “groceries,” “tickets”),
2. And in project 2, the data was explored for finding type of spending lists (think ‘vacations,’ ‘dorm,’ ‘sports team’).

The result

Two scripts/algorithms were eventually delivered, one for clustering and thus categorizing users’ individual expenses, and another for clustering and categorizing entire lists of expenses. In this way, the data has been cleaned up and – for instance – over 50 different terms for toilet paper now use the category “groceries”. The same is true for the lists, instead of ‘Huisje, boompje, beestje’ as the list name it is given the category ‘Household’.  The documentation provided shows how the student arrived at the results, insights were gained into the quality and structure of the data, and follow-up steps to the analysis were provided. This gives Splitser additional insight into their users and triggers follow-up steps.