published on 2020-11-04T10:36:07Z
Algorithms have digitized and automated the lending business in recent years. Nowadays, instead of the clerk, self-learning algorithms do the credit check. But: Such algorithms learned in pre-crisis times. Are your assumptions still valid after Corona? That is the subject of the new edition of the “Digital Quarter Hour”. Ingmar Stupp (CPO of Finiata) discusses how reliably their own customers pay off their due installments even in times of Corona, what hurdles and difficulties there are because the algorithms learned before times of crisis and how quickly they adapt to the new circumstances to adjust. Following on from this, he basically talks about the challenges of machine learning, especially in times of economic and social upheaval. Background: Finiata issues a short-term credit line to self-employed and small businesses. The credit check is carried out exclusively digitally and almost completely automatically by the “Copernicus” algorithm within a few minutes. The failure rate is now two percent, the Gini coefficient, which stands for the quality of the forecast, is almost 70 points, around 30 points above the rate of comparable credit agencies. I In the second quarter (around 80 percent), Finiata’s customers repaid their installments almost as reliably and on time as in the first quarter of the year (85 percent) and at the end of 2019 (84 percent). This at least suggests the assumption that the criteria by which Finiata’s self-learning algorithm “Copernicus” assessed the creditworthiness of customers in times before Corona continue to apply – and that the algorithm does not have to learn from scratch in times of crisis. Ingmar Stupp was responsible for building Finiata’s machine learning algorithm “Copernicus”. He worked as a software developer and CTO.
– .
