Our CEO met with Professor Altman at NYU Stern to get insights on the widely used industry Z Score model while discussing approaches to the VendorMach Trust Score.
While the Z score is designed for public companies, (and our focus is private companies (and SMB’s)), particularly interesting is that considering the era when it was built, it remains a relevant reference point to understand
The Z score looks mostly at company fundamentals (balance sheet, P&L, cash flow statement, income statement), stock prices and then assigns a value in an attempt to predict bankruptcy. Fortunately, there’s millions of historical and timely data to tap to test and improve the model.
With private company scoring, we don’t have the luxury of perfect data to learn from. In fact one of the questions that Altman asked Ojinnaka is how can one rely on speculative big data in modelling?
To that we must first clarify, a few research questions.
What’s the probability that the revenue reported on an SMB tax return is true? Can one predict that revenue with unpaid and paid invoices and the contribution margin much more before hand thus enabling earlier business credit lines?
In the absence of perfect data (annual return and audited financials), is it possible to use a proxy to predict financial standing?
Some other questions to ponder
Is it possible to predict the reputation (trust) standing of a company based on its existing relationships with multiple parties?
With neural networks, a branch of artificial intelligence, we can train our algorithms to become intelligent overtime and predict the future much more reliably. This can be done by scanning hundreds of thousands of invoices, clustering them and training them. The training starts with (aggregation and a regression model on millions of existing credit and bankruptcy data). Overtime, the algorithms begin to get more intelligent, making its own assumptions, relying less on its trainers while detecting new patterns.