Last week I attended the 13th annual banking credit risk management summit in Vienna. It was an amazingly organized event packed with extremely relevant technical and business talks. The audience included mostly people from risk teams in various banks, all with one mission in their minds – improve the credit business of their organizations by pushing the limits of their predictive models.

During the days of the conference I had lengthy discussions with dozens of participants. I heard a lot of the pains we’ve been hearing for the last few years, but also learned a few new things about the industry. In this post, I’d like to highlight just a few of the main takeaways from the summit.

1 – Using Machine Learning for Credit Risk in Regulated Banks

In Europe there is already a small group of banks that got the ECB’s clearance to use machine learning models for credit PD (probability of default) in production models. Many other banks are in the process of getting such approval and the general notion of people I’ve spoken with is that this is quickly becoming the new reality of credit risk modelling. In the past, we’ve written about the many benefits of refreshing the bank’s modelling techniques. Don’t be left behind.

2 – Risk teams still spend too much time on dull tasks

Wow, this is a common complaint by risk experts… Their teams are devoting far too much time for dull tasks which can be tooled, automated or engineered. Lack of proper tooling eventually translates to them investing precious time in software engineering rather than risk modelling. This cry is not specific to the general data pipeline tasks of acquiring data, joining it, cleaning and analyzing it but also to the regulatory requirements of proper documentation and auditing.

3 – Winter is coming

The great recession of 2008 was almost 12 years ago and as well as the economy booming continues, macro economic indexes are signaling a new recession might be on the horizon. While the climate is still positive at the moment, the bank’s P&L owners are starting to push hard on the risk teams to prepare and model for down turns. This creates much more pressure on the risk teams to prepare in advance with more models and quicker ways to refresh their existing models and utilize more fire power to their modeling arsenal.

4 – FinTechs and other tech giants are stepping into credit

It seems that almost every other week there’s a new announcement of another credit providing fintech startup. Lending club just announced the purchase of Radius bank. Sometimes it’s for the underbanked and thin filed population and on other times it’s for small businesses. Due to lower regulatory requirements for these fast moving organizations they are able to take full advantage of cutting edge modeling techniques and tools, often with low visibility and explainability.
If that’s not enough, on more special occasions we hear about another tech giant who enters the credit market making the competition even harder. Just recently Amazon and Apple took some significant steps and are new offering credit. These giant techs come with the infrastructure, tools and personnel to quickly become leaders.

5 – Balanced modelling solution is key

There’s an agreement among the people I’ve discussed with that the right solution for analysts in banks should hold a good balance between usage of novel modelling techniques and enough control on the modelling process. These controls include many of the ingredients which are required anyhow by the regulator and we are all familiar with already: proper paper trail, model books, explainability and anti-biasing and so on. At the end, the solution should both be automated as well as open and transparent, preventing any magical black boxes.

6 – Data utilization, both internal and external

Financial organizations are known for having enormous amounts of data. Using this data in a clever, useful and repeatable way is no easy feat. Often the blocking factor in utilizing internal data for risk teams arises from their dependency on adjacent teams (like data teams, warehouse teams) to prepare data panels for them. Freeing the risk team to explore the available data independently can contribute tremendously to their ability to produce better models quickly.
What hasn’t been said about external data … eventually it’s less about new cool APIs or new data supplying vendors but more about the consumability of this data by the risk teams. Simply put, the inhibitor for the consumption of external data is not availability but rather integration. Tools which will win this integration challenge will pave the way for their users to create much better credit risk models.

7 – Risk teams should become key to the entire model life-cycle

Many people I’ve talked with mentioned that credit risk experts and statisticians are often locked in a center of excellence where their capabilities are only translated through the models they produce. It would make much more sense for the risk team to assume a key role through the entire process, from data, through research ending with deployment. In too many organizations the process is still scattered over numerous teams, technologies and managers. With such a complex approach to the credit risk solution, no wonder that it often takes 18 month to deploy a new model…


Talking with people in our field is for me the best way to fine tune our path as a company. The people I’ve met at the Vienna summit are exactly the audience to which engineer our solution. For many of the challenges this industry is facing, I believe we can offer a compelling solution which will allow credit risk teams to take more independence and authority when it comes to modelling.

If you are also facing some or all of these challenges, I urge you to contact me and hear about our solutions.