RiskMinds International is now only 2 weeks away, and we've already featured several rising risk management professionals as part of our FutureRiskMinds series. We welcome Stefan Heise, Product-Owner Cluster IC at Commerzbank AG, to discuss his thoughts on developments in risk.
What does the future customer expect from a state of the art credit approval banking approach? Customers are increasingly less willing to wait days or hours for their credit approval. Online banking portals educate customers to get quick answers to their requests. Hence, it’s not a matter of time whether complex credit decisions will be granted in real time by smart algorithms; instead the question is rather how fast the classic credit decision processes will be completely replaced by algorithms and how long customers will accept credit decisions with a response time of more than 10 seconds? Is this a creeping process of 10-15 years or will the expectations already have changed in two years? Regardless of future customer needs, risk management in the financial industry will change dramatically.
Future credit approvals will be granted by self-learning algorithms. This means that the entire decision-making process, its execution and the portfolio monitoring will be mostly run automatically.
Leaving behind case by case, rating or scoring based decisions, the future credit approvals will be granted by self-learning algorithms. This means that the entire decision-making process, its execution and the portfolio monitoring will be mostly run automatically. The process includes for example loan request, granting, contract creation, payout, permanent review of decision algorithms and continuous portfolio monitoring with regard to cluster risks, fraud prevention or risk-relevant behavioral analysis. For less complex credit decisions, this already happens today. With each learning iteration this approach can be improved and applied to more complex financial products. Furthermore, this development will reduce the number of classic credit decision makers.
Subsequently a new type of risk manager will emerge connected with according hiring and training needs. Although ideal-typical self-learning algorithms can act autonomously these must be further developed and above all monitored. In the initial period of self-learning credit decisions, the technical rules must be reviewed on a recurring basis. In addition, new decision patterns have to be trained, e. g. for new products. The task of the next generation of risk manager is to deal with a field of tension. On the one hand they have to extend the limits of self-learning algorithms on the other they have to limit the automated decision making process and control it from a risk/return perspective. Because future credit decisions are made 24/7 in real time and expected to occur in higher numbers compared to current situation, the monitoring must work congruently with as little failure incidences as possible. If for example a new pattern of fraud occurs on a weekend it is necessary to recognize and subsequently block automatically. This makes further self-learning algorithms necessary, which monitors the first mentioned decision-making algorithm independently.
The task of the next generation of risk manager is to deal with a field of tension.
An adequate risk management has always been a success factor for the current balance sheet or even for the continued existence of a bank. This principle remains in place despite the current massive changes foreseen for the near future. However, this important area of financial services is already becoming increasingly computerized which means that future risk manager will become a goalkeeper coach for the automated system instead of being the goalkeeper themselves.