The human element, a key element of financial analysis in the credit decision
Human financial analysis (based on reasoning) consists in understanding the situation of a company from information and hypothesis in the reading of the balance sheet, the income statement and the appendices. This allows for example to establish the solvency and profitability of the company in order to make credit decisions.
This approach to analyzing social accounts is called diagnostic. It extrapolates the past to forecast the short and medium term future. It puts forward a reasoning methodology linking observation, understanding of data, financial ratios and different sectorial, economic or company life indicators.
The multiplicity of information used in the credit decision
Throughout the process, the analyst makes the connection between :
- The figures, the knowledge of the company and its sector of activity,
- Solvency, compliance with commitments, payment behaviour, sector positioning, market development prospects, growth drivers, current and future investments of a company.
All this information is provided by the analysis of the figures, the economic environment and the company itself. Thus, by focusing on only certain elements, the credit manager discovers behind each figure, indicator, and information at his disposal, a story in the construction of his financial diagnosis.
It compiles and synthesizes all the information in order to obtain a true "snapshot" of the risks in order to make informed decisions in line with the company's strategy and revenue growth, always within the limits of the credit policy.
The expert system and business rules, an aid to the automation of credit decisions
This human analysis process is sometimes automated by an expert system capable of reproducing cognitive mechanisms. This system simulates reasoning on a set of "business" rules to approach a situation, based on scenarios based on the notions of profitability, solvency and operational performance of the company under study.
This approach is a logical-mathematical aid designed to qualify reasoning sequences from formal logic mechanisms using deductive reasoning. It mimics the expert's judgment in an authorized environment of information, fields of investigation and well-defined rules.
However, if such a system can be synonymous with saving time to have a synthetic vision of the decision, the absence of human intuition on complex or new situations can distort the final decision. Moreover, this approach requires up-to-date and qualified data to make the right decision.
The "scoring application" of machine learning, a statistical prediction tool
Where an expert system will try to reproduce human behavioral decisions by following strict and precise rules, the models based on Artificial Intelligence (AI), such as the failure probability score, is purely statistical.
Scoring is a machine learning application built on a predictive probability model. This data science is built around the techniques of correlation (measuring the strength of a relationship between two variables), regression (analyzing the relationship of one variable to one or more others) or dependence and causality of simple or aggregated data.
For credit decisions, this is a real revolution. Of course, there is always an element of uncertainty in these algorithms that we are forced to accept in the automation of decisions. Moving towards Artificial Intelligence is an invitation to reflect on a new rationality in the prediction of behavior.
It is a way to automate in real time the credit decision on large third party portfolios. This predictive approach allows the credit manager to manage his third parties with more security and accuracy despite incomplete data. This allows him to focus on companies that require a more detailed follow-up.
Eliminating bias is at the heart of the benefits of machine learning
Machine learning also has the advantage of eliminating human behavioral and cognitive biases by depriving human intervention of false interpretations in failure probabilities. However, enough data must be available in the creation, development and maintenance of predictive models to calibrate the default probability curves to the highest level for risk classification.
In addition, machine learning applied to poor quality data can also lead to bias. It is often better to spend more energy in collecting and cleaning data than in creating very complex models. Artificial Intelligence is therefore a valuable tool, provided it is judiciously used in the risk assessment chain to predict failure.
Human financial analysis, expert system and scoring, complementary tools in credit management
In conclusion, by comparing these three methods of credit risk analysis, we can see that they complement each other in the decision-making process. With the digitization of processes and the advent of Big Data, the credit decision is at a crossroads between anticipating late payment, non-payment and default. Thus, technologies are transforming the credit manager's job. Faced with the multiplicity of his tasks, he will now be able to concentrate on the most litigious companies.
Today, the added value of the credit manager lies in the critical examination of information both internal and external to the company under study. The assessment elements can be varied, even complex, but their often imperfect understanding very often provides useful answers or explanations for the credit decision.
In conclusion, regardless of the method used, the evaluation of customer risk remains in the hands of the credit manager with a view to preserving cash flow. The credit manager can grant or refuse an inter-company loan at the beginning of the relationship, reduce, maintain or increase an outstanding loan, continue or terminate a business relationship with certain existing customers with high risk profiles. The credit manager's daily tasks include setting payment deadlines and schedules, managing late payments, limiting the risk of non-payment, organizing dunning procedures and handling litigation.
| Type of information | |
|
Factual elements that can easily be used by AI |
|
|
Intangible elements that are difficult to value with AI |
|
