Ga direct naar inhoud

The future credit risk model landscape: is AI the next big thing, or should we expect other trends?


For example, AI could improve the KYC process (if you’re interested, you can check out this blog). But is this also the case within credit risk management? Will regulators allow the use of relatively new technologies like machine learning in credit risk models? Is there added value of AI within risk? In this blog, we discuss different applications of new technologies within credit risk and their possible implications.

AI & data

One of the fields related to credit risk in which AI may become standard is data. These days, banks realize that the quality of their data is as important, if not more, to the quality of their credit risk models. Even if models work perfectly, the results are unreliable if data quality is not up to standard. TRIM findings suggest that many institutions face issues in data quality and data management [1].

So how can AI improve these areas? Machine learning algorithms can detect anomalies in data. The algorithms can find potentially strange data entries that need more investigation. AI technologies can also be used for the automation of report generation, for example data quality or monitoring reports. In this way, employees can spend more time on other meaningful tasks. Another development that we are seeing is that the data and modelling departments are working, and probably will continue to keep working more closely together than they did before., Previously the modelling department prepared the data-sets they needed themselves, but now that the data quality plays a more important role, the data department has taken over. The data department has more knowledge about the data and can prepare a data-set more thoroughly and deliver a higher quality data-set.

AI & credit risk models

Since this is a new field, there are not yet any established best practices regarding the use of AI in models. Machine learning algorithms can become part of PD (Probability of Default), LGD (Loss Given Default) or EAD (Exposure At Default) models, because they can find relationships that traditional methods cannot.

Initially, banks may start with small projects that investigate how AI can be used to improve credit risk models, independent of regular modelling process. Banks can use the same modelling steps, control framework, and role of model validation and hence treat AI as normal AIRB models.  This will also show supervisors that AI models are not just an experiment.

Another option is when banks develop AI models parallel to the current “normal” AIRB models and use them for insights instead of reporting purposes. This way, banks will be able to show supervisors the value additions of their AI models over ‘normal’ AIRB models. Also, both banks and supervisors will gain more knowledge about AI models and warm up to their use. To me, this seems to be the most likely way for AI credit risk models to be used in practice. However, I also see the probability that AI models will not be taken into use any time soon, because banks simply do not have the time and resources to start developing them and or are busy enough with redeveloping their ‘normal’ AIRB models, as well as implementing new regulations like Basel IV.

The credit risk model landscape is heavily regulated. Supervisors are possibly not keen on allowing unknown modelling methods since they are apprehensive of the so-called ‘black box’ algorithms. I think that supervisors will start allowing more complicated AI models gradually, once both banks and supervisors have more experience with them.

New risk drivers

Credit risk models will witness a new type of risk drivers.  It is possible that AI technologies like RPA and NLP will be used to gather the necessary information for these risk drivers. If supervisors are not keen on allowing the use of AI technologies in the risk field, it is likely in my opinion that new types of risk drivers will become the primary focus in term of innovation of the modelling teams.

  • Social media

Social media can reveal a lot of information about a client. For example, if a client has a loan with a car as collateral, and the client meets with an accident in which the car is completely damaged, the collateral value decreases to zero increasing the risk of the bank. Accordingly, the risk rating of the client should be adjusted, and this can be done faster than usual, with the help of social media. Another example is when standard ratings from rating agencies for corporates are optimized by the rating based on data from sentiment analyses from news and social media. However, it should be investigated how privacy regulations like GDPR will affect the use of social media information in credit risk models.

  • Sustainability

Sustainability can impact a clients’ risk ratings. I think sustainability will be a suitable risk driver in business loans models, or for mortgages, to assess the environmental sustainability of the collateral. Environmentally unsustainable companies have historically performed very well. However, looking into the future, it is likely that their risk rating will get worse. Sustainable companies might avoid expensive crises, as opposed to unsustainable companies. In addition, being a socially sustainable company might implicitly contribute to a better risk rating. [2]

In addition, socially sustainable companies are more likely to attract investors, resulting in a better risk rating. A risk driver related to sustainability might be less privacy sensitive, but it might be harder to assess how sustainable a business or collateral is. Nevertheless, it will be more attractive for clients to get a mortgage when the collateral is more sustainable. Consequently, it will be harder for unsustainable businesses to get a loan. You can find information on sustainability modeling in this blog.


I think that AI can bring a lot of value to the field of credit risk modelling. Besides all the advantages, the question remains if regulators will allow innovation by use of AI in the risk field and even if they do, it might take some time.

However, if banks want to keep innovating, it is important to start thinking and acting upon the new developments as quickly as possible. Once AI becomes the standard in all other sectors, banks would not want to fall behind the others. For banks the next great task is to implement machine learning and make AI auditable.