Credit Risk and Machine Learning Concepts — 1
I have been key in implementing a Credit rating and risk system more than a few times, originally considering an implementation of essentially Z-test scoring on cash availability to predict ability to pay bills and remain solvent for UK Business Entities and to some extent basing on similar ratios and inputs from commercial credit rating agencies some international and US credit worthiness. In future blogs I will be going deeper into what this means.
With the advent of more effective and less technical and academic knowledge to implement and leverage the power of Artificial Intelligence and Machine Learning, it would be a good thing to share and provoke my observations and assessment of Leading Practices in this domain.
The essential problem is what business terms can I risk with my accounts receivables and to what extent can I finance my customers (and, in some cases, my suppliers/vendors) in a Supply Chain organization so as to protect my exposure to risk or default on payments and promote rapid cash flow payments as well as loyal and continued business from my Customers and potential customers?
In retail mode, I generally get payment at the point of sale, so some of the activities are not necessary. This is also normal for Business to Consumer, in that the payment for goods or services are relatively simple and if not paid at point which title to an item of value changes from business to consumer, additional finance may be offered through Financial Credit either in house or through Financial Services that will base the credit terms on a personal credit rating and evaluation and score. The risks and exposures to open invoices are minimal, and it is also relatively simple to recover the item of value or pursue civil debt payments, without having a major concern of potential write-offs.
In Business to Business and wholesale or bulk re-sale mode, it is common for businesses to extend credit terms that delay payment of goods at the point of transfer to a future date, set out in contracts of sale or purchase and sale agreements. A small business may set a 15 days net on invoice date as a standard term, but some large customers may have reasons and in fact prefer to protect their own cash flow by negotiating longer payment terms. Trade Credit is commonly expected, in that, at the moment the title of ownership of a thing or service of value changes from the supplier to the business customer, there is a period between a receipt of the item, the generation of an invoice for payment and the receipt of the consideration to meet the obligations of the transaction by the Business customer and recipient of the item or service.
The age-old question in every Credit and Risk Manager in a corporate finance department is ‘How to gauge the credit-worthiness of a trading partner?’
There are 3 approaches commonly, and the problem of risk evaluation can be decomposed into 14 topics and sub-topics.
Approach 1: Entrust all or a subset of sales invoices to be covered by a commercial credit insurer or Finance Refactor of outstanding invoices
Approach 2: Use a commercial credit reporting agency to manage credit ratings and assume all risks for every sale transaction.
Approach 3: Use a combination of Commercial Credit Agency reporting, past trading history as an existing customer, non-structured external data with a mixture of rule-based calculations and Assisted Intelligence and Deep Learning models. In future blogs, I will detail some high-level insights into what leading practice and my current activities in this area are finding.
So, with all these approaches, the point must be that risk is managed and very few, if any, outstanding customer invoices have to be written off.
Approach 1 has the advantages that at least 80% of the invoice amount is typically guaranteed, up to a maximum for any one Customer in the situation where a Commercial Credit Insurance agency is used.
Approach 2 is fine for stable organizations as your customer, or more localized and smaller businesses. I will discuss in the next blog my activities to model a similar credit score, which I did before online commercial credit rating and the internet was a thing!
Approach 3 will be discussed in a future article.
The next installment is available here :
