AI in Credit Management

The field of artificial intelligence is very broad. It includes rule-based systems, so-called expert systems, which have long been used to control processes or make decisions in credit management.

In this blog post, we refer to methods based on supervised or unsupervised learning. The basic principle here is that we need large amounts of data. However, if we are in B2B, then often, at least for the applying companies, they are too small to really use AI. In most cases, there is a lack of a sufficient number of failed companies for an insolvency forecast, for example. So, what then?

Sufficient data exists more often in payment experiences, for example. Here, one could try to quickly identify bad payers not only based on certain patterns. Based on payment behavior, an attempt could be made to identify how the risk of non-payment at the company in question is changing. This in turn could be included in a company rating.

Another aspect where we are really making progress with AI is text analytics. This is where information about a company from different sources which can be found on the Internet can possibly indicate that a company is experiencing economic difficulties and potentially payment difficulties. In the first steps, it can also certainly only be a "warning system" for the credit manager, which then needs to be verified by personnel.

If one has extensive quantitative payment time series of the regular customers, one could perhaps also make a better forecast of incoming payments than before. However, this task is extremely complex and requires further in-depth research before truly satisfactory solutions can be presented here.

If one has extensive databases available, then AI can support, for example, where different types of data are linked together. One currently interesting research topic is the question of whether classical financial statement analyses can be combined with the automatic analysis of text-based information, e.g., from appendices or analyst reports, in order to achieve better forecasting results. Here, a wide range of information can be considered in a complementary way. However, the improvements will be gradual rather than revolutionary.

Wherever flexible business processes exist in which the result depends on the path taken, patterns could also be included if there is sufficient data to identify which path should be taken and when. A typical example of this is the customer approach when demanding outstanding payments. What is the most successful approach?

In the future, it may be possible to use a customer or prospect profile to infer which payment methods should be offered or which hedging strategies should be adopted for the business in question.

In the case of private customers, on the other hand, there are usually far more cases and data available. Here, classic customer ratings can be enriched, for example, by the activities of individuals in social media.

In the analysis of numerical data, there are now procedures in which rapid analyses are generated with AI-based solutions. For regular operations, these are replicated using classical statistical methods.

In summary, these approaches in B2B business will certainly bring improvements in various places. However, we are not talking about disruptive changes, but rather gradual improvements. In this respect, the expertise of the credit decision maker will still play a decisive role in the future. With this in mind, I wish you good decisions in these turbulent times.