To say that Generative AI (Gen AI) is game-changing is to state the obvious. To put it in a historical context, Gen AI is one of a few significant general purpose technologies (GPTs) that have provided society with enormous productivity gains over the last few centuries.
GPTs such as the steam engine, electricity, the semiconductor and the internet follow similar life cycles. They begin with an initial productivity boom driven largely by a substitution effect, followed by a much greater augmentation phase propelled by massive downstream innovation.
With Gen AI today, we鈥檙e still at the beginning.
Gen AI is still in the substitution phase, which brings a certain amount of anxiety associated with the so-called 鈥淭uring Trap鈥濃�攖he fear of machines replacing humans. But this is also not without historical precedent. The same anxieties drove the Luddites to smash factories 200 years ago in the UK due to fears that the steam engine and industrial revolution would take jobs away.
But we鈥檙e already witnessing real productivity gains. Intelligent automation can now provide the benefits of Adam Smith鈥檚 鈥渄ivision of labor鈥� in a mixed world of digital and human workers. Digital agents can take on many routine tasks that are highly repetitive and time-consuming, leaving humans to focus on higher-level tasks that require true judgement, interpretation and analysis.
Illustrative example: Credit Risk
Consider the credit risk function of a large global bank as an example. Typically, a credit limit hierarchy is in place with limits set at various exposure levels, such as counterparty, asset type, industry sector and jurisdiction. Excess management is the process that manages the workflow around the bank鈥檚 response to each breach of a limit. For a typical bank, there could be thousands of breaches in a given day where some form of intervention is required. Furthermore, upwards of 80% of those breaches can be 鈥渢echnical鈥濃�攅xcesses caused by input mistakes, bad data or reconciliation errors.
So why could this be a big problem? If every breach is a breach, humans spend far too much time dealing with issues that are not material, which is an inefficient use of skilled resources and, in the worst case, can lead to unnecessary credit line freezes impacting client business.
So, while all breaches require some form of intervention, digital workers are better suited to address the technical ones because they are quicker, more efficient and run during off-hours. Tasks include categorizing breaches by type and status and acting on those that can be resolved mechanically, leaving humans to focus on proactively addressing true material breaches.
The substitution benefits of Gen AI already provide significant dividends to society, and the next phase is augmentation, which will be the biggest driver of productivity gains鈥攃reating a feedback loop that stems from extensive innovation.
Generative AI and risk management
The challenge for financial institutions in a world of increasing geopolitical, economic and environmental volatility is to consistently generate Economic Value Added (EVA). On the one hand, this means innovating and creating positive business benefits as an 鈥渙ffensive鈥� pursuit while, on the other hand, controlling the costs of doing business, GRC, infrastructure, capital, funding, etc., as a 鈥渄efensive鈥� pursuit.
The combination of Gen AI and intelligent automation provides an unprecedented opportunity for financial institutions to navigate both pursuits holistically. Gen AI provides tools to analyze and generate new content in various forms, creating unexpected insights and value for firms, while intelligent automation provides business process management (BPM) and robotic process automation (RPA) capabilities to execute on those insights.
Financial risk management has yet to significantly benefit from the promise of Gen AI. The focus of the risk function in most banks has been driven primarily by defensive objectives to address the regulatory reporting requirements of Basel, which initially focused almost exclusively on capital adequacy. To the extent that large language models (LLMs) have been leveraged to wade through regulatory updates, and machine learning models have helped dramatically speed up simulation performance, Gen AI has already played a role in reducing infrastructure and overhead costs associated with the risk management function.
On the offensive side, however, the realized benefit of Gen AI has lagged. Up until the 2008 global financial crisis (GFC), a stated goal of the ongoing evolution of regulation was a convergence of 鈥渞egulatory鈥� capital and 鈥渆conomic鈥� capital. This philosophical approach motivated the use of risk measures such as portfolio volatility and Value-at-Risk (VaR) for reporting purposes鈥攎etrics drawn from modern portfolio theory (MPT). The idea is that enabling banks to align regulatory reporting with best practice risk management would also better align a bank鈥檚 defensive pursuit with the offensive pursuit of doing better risk-adjusted business.
The GFC changed everything. Regulators quickly abandoned their goal of converging regulatory capital and economic capital, and began to move beyond the singular focus of capital adequacy in attempting to restore stability in the financial system. At the same time, the GFC also served as a wake-up call for best practice that had already been evolving away from its comfortable adherence to MPT and the efficient market hypothesis (EMH). New paradigms such as behavioral finance, mental accounting and irrational exuberance had already gained prominence in academia and the industry.
Even if it can be argued that there is still much value to be derived from the principles of MPT, its implementation in practice has always been problematic. The premise that a time series of asset returns or risk factors provides all the information required to determine possible future distributions has consistently failed in practice for various reasons. Empirical challenges have included the twin curses of non-stationarity and dimensionality, as well as the overarching assumption of market efficiency鈥攚hether it be strong form, semi-strong form or even weak form.
It's here where Gen AI can be truly disruptive and transform financial risk analytics, and also change how risk groups engage with the front office, supporting trading and portfolio optimization.
By bringing Gen AI into the picture, one can start accessing much broader sources of data鈥攇oing beyond the use of very structured time series data to incorporating unstructured data sources and real-time information drawn from market sentiment, social media and trading patterns. While this requires much greater volume and heterogeneity of data to process and draw patterns from, it is precisely what Gen AI does best.
The future of financial risk management
As risk managers increasingly incorporate deep learning and reinforcement learning techniques into their risk analyses, they can perform much more informed scenario analysis, gaining greater insights into predicting possible future outcomes. Importantly, this will enable risk groups to provide significantly greater value to the front office through improved hedging and portfolio optimization models, utilizing synthetic data to implement increasingly popular techniques such as deep hedging.
Fundamentally, this will allow financial institutions to be much better at anticipating and managing low probability but high impact events鈥攁nd not just the normal market shocks鈥攖hat are the greatest cause of concern.
Contact us to discuss how your firm can leverage the power of Gen AI and intelligent automation.
