1. Customer onboarding and Know Your Customer (KYC)
The KYC process is primarily a regulatory obligation imposed on banks and financial service providers to prevent money laundering and terrorist financing. Beyond that, it offers the opportunity to create a comprehensive customer profile that, if properly maintained, provides all relevant information needed for regular sanction list and PEP screening, or for periodic updating of the credit rating, for instance.
Especially in the KYC process, together with onboarding, the potential efficiency can be raised considerably by means of digitisation and automation. Solutions like ACTICO KYC can be integrated into the onboarding process via suitable interfaces and, for example, take over automated comparison of customer data with sanction lists and PEP (politically exposed person) lists, updating of risk classification, or documentation of a company’s beneficial owners. The KYC profile can be updated according to risk. In this way, a solid data pool is created, which meets the requirements for all aspects of credit management.
2. Creditworthiness assessment
The basis for assessing a company’s creditworthiness is balance sheet analysis. Annual financial statements and quarterly reports do provide extensive data on a company’s financial situation, but the acquisition and analysis of this data is often quite an obstacle. Slow manual processes delay credit decisions and increase costs.
The use of artificial intelligence (AI) can automate the incorporation and reading of balance sheets. With automated spreading, financial data is captured from financial statements and assigned to the appropriate categories. This means that all customers’ data is available in one uniform format and can easily be processed further.
However, a complete overview of the financial situation can only be obtained by adding data from various internal and external systems, such as credit agencies, along with qualitative information – from social media, for instance. Natural language processing (NLP), a branch of artificial intelligence (AI) like ML, can evaluate a customer’s reactions in forums, for example, and thus identify positive or negative changes to their image. For instance, frequent complaints about quality defects could be a warning signal regarding a company’s future financial development. Such automated processes can be used to acquire much more data about a company than before – and thus paint a more complete picture that minimises uncertainties in risk assessment.
3. Risk quantification
Risk quantification comprises determining the probability of default (PD), loss given default (LGD) and risk-adjusted return on capital (RAROC). It provides the basis for pricing and other credit terms.
Commercial lending today is still largely based on a manual process involving in-depth analysis of baseline data and evaluation of soft factors. In such cases, the decision depends, to no small extent, on the loan officer’s experience, which has a relevant influence on the weighting of the various risk items. This is not necessarily a disadvantage: The human factor can be a corrective when it comes to data from manipulated books or unrealistic forecasts regarding sales and growth. On the other hand, this can allow systemic errors to creep in, with a negative impact on the lender’s margin. One study of corporate loans in Portugal concluded that, in 2018 and 2019, especially for longer-term loans, the risk-adjusted return was almost 3 percent too low. Sector-specific differences were also identified. For example, companies in the construction, real estate, transportation and logistics sectors received their loans with conditions that underestimated the actual risk, while the manufacturing, wholesale and retail sectors tended to receive overpriced loans.
Latent mistrust of self-reported borrower data has resulted in a mesh of manual processes and cross-checks that are still in use today, because trust in these methods is higher than in modern technologies. However, scientific studies show this to be ill-founded.
The white paper ‘Machine Learning in Credit Risk’ from Banco de España explains that the use of AI or ML algorithms in risk modelling leads to predictions that are up to 20 percent better. One additional benefit of ML models is that they not only calculate the borrower’s ability to repay, but can also evaluate other price-relevant criteria. For example, there are applications that also take behaviour into account, such as observed loyalty towards the institution or price elasticity with regard to cross-selling strategies, in order to find the optimal segmentation. The positive effects for the bank are reflected in reduced losses, better or more favourable capital requirements and lower operating costs.
4. Credit decision
It is true that banks can currently look forward to greater interest in asset financing, but in view of the ever-shorter innovation cycles and volatile development of the German economy, it is also the case that innovation decisions often have to be made at shorter notice today than ten years ago. On top of that, queries are becoming increasingly individual and complex – in other words, more time-consuming. Nevertheless, prospective borrowers are rarely willing to accept the resulting longer processing times and higher costs.
Acceleration of credit approvals by means of automated, more efficient processes is therefore an essential competitive factor. One example of how this can be achieved is shown by the introduction of ACTICO Credit Risk Management Platform at SüdLeasing. This financier uses an automated initial check, based on 25 risk-relevant rules and a decision engine. Detailed customer data is incorporated by importing electronically structured annual financial statements. This is followed by automatic calculation of the debt servicing ability, the rating and the property risk, in accordance with the rule models developed by SüdLeasing. From this, the application generates a decision recommendation.
In this way, 80 percent of credit requests can now be decided on within just 48 hours. For more complex cases, which previously took up to six weeks, SüdLeasing has been able to reduce the processing time to an average of two weeks.
Read Success Story
5. Price calculation
When calculating credit terms, many banks still rely on a ‘one size fits all’ approach, which can only be deviated from within narrow limits. As a result, creditworthy customers have to pay a premium to subsidise riskier customers.
Meanwhile, machine learning has already become the tool of choice for pricing a wide range of financial products. It can also be put to more use in the lending business. It allows the individual probability of default and the general repayment performance of the borrower to be determined very reliably. For banks and financiers, this presents an opportunity to depart from the old rigid pricing scheme and switch to dynamic risk-based pricing.
Alongside the more attractive conditions, a transparent presentation of the considered factors and, on this basis, an individual offer that corresponds to the actual risks can increase the customer’s trust in their bank and thus make a further contribution to strong customer retention for the long term.
6. Monitoring after payout
As long as the borrower pays their instalments on time, everything is fine. However, if problems arise at some point, it may already be too late. It is therefore vital that banks also monitor the borrower’s ongoing development, so as to be able to react to changes in a timely manner.
If, in addition to the balance sheet figures, qualitative data on the company is also acquired, then conclusions about further development can be drawn at an early stage. ACTICO customers, for example, use machine learning procedures to calculate the probability of delayed payment. This already occurs in the context of credit card customers, for instance. In the case of corporate loans, decisions about extending the credit line, for example, can be made at short notice on the basis of this risk assessment.