The business – Swishfund provides short-term loans to entrepreneurs. Based on their bank transactions, Swishfund analyses the cash flow of a potential new customer. Within 24 hours, entrepreneurs receive their loan proposal.
The challenge – The ambition of Swishfund is to scale quickly, completely automating the customer onboarding and risk evaluation process becomes key. The Swishfund data science team already developed rulebased models to analyse new customers. The challenge was to escalate these models and the underlying infrastructure to a higher level and bring the whole workflow to a production state.
The solution – Together with the data science team at Swishfund, we developed a machine learning platform. On this platform, their first 3 ML models run in production. ML models predict risk measures and determine the loan value and terms. Swishfund experts constantly evaluate these data-driven proposals. A feedback loop feeds the models in order to learn and improve. In this way, the knowledge of the experts can be incorporated in the models, avoiding black boxes.
The technology – The Swishfund Machine Learning platform was realized by using open-source software and technologies. ML models are built in Python, and the data science platform uses Airflow, Docker, Flask, running on AWS. Enjins helps scale-ups like Swishfund.