The availability of bank statement data via companies such as Yodlee has resulted in a boom in FinTec companies. Products derived from this data include, financial management, wealth management, fraud prevention, marketing and accounting.
One such product was developed by my team in a previous company, the aim of which was to allow companies to benchmark their sales against competitors. The product took the data feeds from a data provider who had enriched each transaction with a merchant and category, loaded into an MS SQL data warehouse and then aggregated and published via Tableau.
The product had limited appeal due to the lack of demographic data. Fintec data would on the face of it appear ideal for a machine learning approach. Using engineered features and clustering algorithms it should be possible to create the user segmentations needed by companies to slice and dice the data in order to make an effective marketing tool.
This repository contains a Juypter notebook with my investigation into the plausibility of such an approach and a brief report highlighting my findings
NOTE: due to the size of the notebook it can take a couple of attempts to render on github.