Without risk, there are no rewards. Financial institutions, by their very nature, are risk-taking entities. They have to balance the risk against the potential benefits for the good of their investors, shareholders and members.
Today, banks and financial institutions are deploying model risk management (MRM) solutions that are more sophisticated and varied than ever before. The number of models is also rising dramatically, according to analyst McKinsey – 10 per cent to 25 per cent annually at large institutions – as banks utilise models for an ever-widening scope of decision making. More complex models are being created with advanced analytics techniques, such as machine learning, to achieve higher performance standards.
An increasing reliance on models, regulatory challenges, and talent scarcity are driving banks toward a model risk management framework that’s both more effective and value-centric.
A model-driven architecture
The more sophisticated the model, the greater the chance of introducing errors into the decision-making process. Risk governance must tackle a number of issues, including the challenge of contamination or interconnected risk between models. There are always concerns around data, not just the quality of data used for developing the models but also whether the correct data is being used when the model is running.
Risk models themselves can be poorly used or exploited for the wrong purposes, such as fraud. There are a number of examples of risk models leading to large financial losses. Many have occurred due to manual data input and basic human error.
The solution is to turn to artificial intelligence (AI) to reduce these risks. In software pioneer Tom Siebel’s book, Digital Transformation and Mass Extinction, he recommends companies stop the ‘do it yourself’ approach to digitisation. Instead, successful organisations should build their applications on a tried and tested, pre-existing technology stack based on a model-driven architecture.
This architecture relies on the confluence of several different technologies: elastic cloud, big data, AI and the Internet of Things (IoT). Digital transformation is no longer simply about improving operations. Instead, it will disrupt business models to their core and change how we think.
At the heart of this transformation is AI. Companies that invest in the technology often go on to create labs for experimentation and become real centres of innovation. They attract specialist and expert personnel while creating a new infrastructure for their digital business models as a result of the potential it creates.
Indeed, our research shows more than eight out of ten (81 per cent) adopters of AI are seeing benefits. The greatest performance improvements expected from AI over the next three years include gaining faster insights from data (78 per cent), reducing manual tasks (77 per cent) and improving decision-making (77 per cent).
On the rates of adoption by risk use case, companies were expecting process automation to be the biggest AI-related improvement. Credit scoring, data cleansing and enhancements, and risk grading weren’t far behind.
Facing down the challenges
With financial institutions facing a battle for skilled MRM-capable staff, it makes sense to adopt an AI strategy that automates the most basic processes. AI can assist with automated model documentation, populating validation documents from the model inventory, and in the near-future natural language processing will allow additional insights to be included in textual form.
AI can also provide an automated model usage logging system and identify areas where models have been used for which they are not validated, or indeed identify models that are no longer being used.
There’s also huge potential to create automated alerts for identifying where models are failing to perform or where they shouldn’t be used due to a change in the environment. So, if a model tracks interest rates, it can trigger an alert if they rise above a certain level - a risk factor that would otherwise be missed amongst a mass of data.
Machine learning models require more sophisticated techniques than automation and much more data. Potentially, however, the benefits are also much greater. One of the obvious benefits is for text mining of documents to look for certain keywords or phrases. It’s a speed reading and labour saving method of scanning complicated and weighty documents. The machine learning tool can learn certain phrases that are of interest, and flag models that need additional validation or are of priority for internal audit.
Over time, machine learning can learn about what is a ‘good’ and ‘bad’ model and flag that up for checking. It’s not a fool-proof method and still needs the human check, but the time-saving potential is huge.
Investment worth the reward
In order to benefit from AI and Machine Learning, investment is needed to provide the appropriate environment, including access to big data and associated processing power, and the ability to deploy them. Providing robust model risk governance is also key in overcoming reluctance to allow the use of AI and ML models.
Using these types of models to improve MRM has additional benefits. Automating tasks allows a resource constrained team to do more in a shorter time and be well positioned to deal with the expected increase in models. Using machine learning can identify potential for model risk and flag areas of priority.
With that comes additional costs for data storage, processing and training. Without additional investment across all these areas, the benefits of AI just can’t be realised.
At the end of the day, however, the benefits outweigh the costs. Embedding AI tools in the MRM equation are meant to be enhancements, not replacements for human effort. The tools can produce massive efficiency savings and reduce error, so the potential time, efficiency and productivity benefits are well worth that extra investment.
Dr Iain Brown, Head of Data Science, SAS UK & Ireland