Doing data science for SME financing often feels similar to prospecting: Instead of tons of gravel, the data scientist sifts through tons of data (credit applications, accounting data, books of receivables) in search of valuable nuggets – i.e. the low-risk suppliers to lend to.
But what if, at least here, the Philosopher’s Stone was more than a myth? What if it was actually possible to transform the SME’s risks by re-focusing the analysis? For thousands of SME suppliers trading with large corporate buyers, this can be done.
An invoice by a small, credit-challenged supplier may have a terrible supplier risk, but it could have a very high probability of being settled. If the buyer is a highly rated, multinational corporate buyer, this invoice is a very attractive risk to finance to the benefit of the supplier. The important question is the buyer’s likelihood to settle the invoice, not the SME supplier’s credit risk.
Accessing this opportunity requires, however, the ability to accurately measure the probability that a given invoice will be settled by the buyer. This can be done by leveraging the data in the ERP systems of the large corporate buyers and training a state-of-the art ML-based scoring algorithm on it. This algorithm produces a “probability of settlement” for any invoice, without any limitations to scale or scope.
Hence, the advent of modern machine learning has made possible in the field of SME financing what medieval alchemists could only dream of: The transformation of leaden SME risk into golden investment-grade large-corporate risk. However, this transmutation required a technology that would have been inconceivable to a medieval alchemist.