top of page

Navigating a fine line – How we balance the risks and rewards of generative AI in commercial portfolio risk monitoring


What an 18 months generative AI has had! Like an epic coming-of-age story, it went from being an image-generating sidekick to virtuoso hero tech in no time at all. In just 5 days of its launch, ChatGPT amassed 1 million users. Unsurprisingly, businesses were quick to jump on the bandwagon with every company from Adobe to Zoom quickly releasing their own suite of generative-AI-based tools. Unfortunately, it hasn’t been all smooth sailing for the tech. Generative AI comes with its own set of ethical, legal, operational, and regulatory risks. Left unchecked, these could undo some of the benefits of using such systems. Here’s a brief look at the good and the bad of generative AI and how we at TRaiCE navigate the fine line between harnessing its rewards and mitigating its risks.  


Person using a generative AI tool on a laptop
Generative AI can be a game-changer in a data-driven domain like commercial portfolio risk monitoring

Understanding the risks and rewards of Generative AI in commercial portfolio risk monitoring


The good


Traditional commercial portfolio risk monitoring is an inherently inefficient process. The reviews are often irregular, dependent on past data, and involve manual oversight. This leaves room for human error and dangerous risk blind spots – a problem that compounds as the size of your lending portfolio increases. Generative AI can process trillions of data points (both structured and unstructured) within seconds. It can therefore be a game-changer in an information-driven domain like commercial portfolio risk monitoring.    


One of the key benefits for risk managers is generative AI’s ability to analyze all kinds of data and uncover valuable insights and intricate patterns from it that could point to future risk probabilities. The inclusion of more data points automatically gives way to improved monitoring and more accurate risk predictions helping you make timely, data-backed decisions that enhance your portfolio’s performance.


Additionally, with the infusion of current financial, non-financial, and market data, AI systems can analyze portfolios in real time, giving you the ability to stay agile and finetune your strategies promptly even in volatile market conditions. The fact that AI can do this for you day in, and day out, no matter the size or complexity of your portfolio is of course the icing on the cake as it provides ongoing monitoring while freeing you up to focus on other strategic tasks. According to McKinsey, generative AI can improve productivity by 70% or more resulting in a value add of trillions of dollars annually to the global economy. 


In short, generative AI can be to commercial portfolio risk monitoring what washing machines are to households – they get more done, in less time, repeatedly, and with better quality so you can be more productive in other areas.


The bad


For all its immense capabilities, generative AI can sometimes display the behavior of a 2-year-old, in that, it can spout utter nonsense with utmost confidence. This type of behavior where the system generates false or misleading information is called an AI hallucination. The ramifications of such behavior can be significant in the risk monitoring domain if the AI system is taken at its word. For example, if an AI system inaccurately predicts risk based on market trends that aren’t real, it could lead to lending decisions that result in substantial losses.


AI systems may also make spurious correlations mistakenly linking unrelated variables with creditworthiness. For example, an AI system might correlate the weather pattern in a region with a company’s creditworthiness even if there is no meaningful relationship between the two. This can lead to the system misclassifying a healthy business as a high-risk borrower, resulting in managers denying them a loan renewal or imposing higher interest rates without justification.


There are also broader issues of compliance and transparency. Complex generative AI systems can have several decision-making steps, some of which may even be hidden. The convoluted processing calls the explainability of the final AI output into question, which in turn makes it hard to ascertain if its assessments are fair or biased. Such opacity only serves to erode trust in an industry already riddled with allegations of biased lending. This along with security and privacy risks increases the potential of regulatory scrutiny on generative AI systems.


Striking a balance


At TRaiCE, we have several checks and balances in place to ensure that our users get all the capabilities of generative AI without its less desirable facets. Here’s how we do it:


Keeping it real


In the finance world, facts hold significance while creative fabrications are frowned upon. So, why should it be any different for an AI-led credit risk monitoring system? Accordingly, our AI model is ‘tempered’ such that it sticks to what it has learned and what it finds in the data it is fed. In technical terms, this involves adjusting the model’s probability distribution such that the outputs it generates are fact-based and not creative fabrications, thereby reducing the likelihood of it hallucinating.


In addition, we maintain an information-dense environment. Our algorithms collect data from over 70,000 quality web resources including reputed media sources, credit bureau reports, corporate records, regulatory documents, legal filings, etc. We regularly monitor, validate, and cleanse these sources to ensure the accuracy of our system’s risk assessments.     


Adding context and relevance


One of the keys to decoding information properly is correctly judging the context in which it is set. For example, interpreting a text message without knowing the sender's mood can easily lead to misunderstandings. The same holds for AI systems which rely mainly on the generic context of the text to generate an output. This imprecise methodology is what leads to systems making illogical correlations and inaccurate assessments.


At TRaiCE, we avoid this by using carefully crafted prompts, meta data, and fine-tuning our models by explicitly training them to identify financial contexts within the text. In layman’s terms, this means that our system is trained to not only identify default-indicating red flags but also the relevance and impact that this data will have on a business’s creditworthiness correctly.  Furthermore, we allow our users to provide additional context and/or corrections. This feedback loop essentially allows TRaiCE to iteratively improve its risk assessments over time.


Explaining everything


TRaiCE is what industry experts call a ‘glass box model’. Its processes are fully auditable, allowing users to see all the metrics that the system takes into consideration when assigning a risk score to an entity. And if you can trace it, you can also finetune it. This has three benefits. First, it ensures regulatory compliance and allows lenders to weed out any bias-inducing data. Second, it helps to identify and correct AI hallucinations. Third, it gives you the ability to make the system more targeted and accurate. For example, for monitoring real-estate portfolios, you can finetune the system to identify sharp declines in property values or new zoning regulations in a specific market. This makes TRaiCE a highly customizable and accurate risk-monitoring AI tool.  


Democratizing tech access


Generative AI systems can be costly to implement. Most businesses lack the in-house skills to implement a bespoke intelligent system that can process huge amounts of data. Outsourcing the task can also be an expensive affair. As a result, only top-tier financial organizations have ready access to such systems and their benefits. We're on a mission to bridge this technology gap, making the benefits of these tools affordable and available to a broader spectrum of financial institutions.  By giving more FIs access to real-time risk intelligence, we believe we are empowering and encouraging the industry as a whole to grow.


A banner encouraging people to schedule a demo

Conclusion


At TRaiCE, we’re excited about all that generative AI has to offer to the world of commercial portfolio risk monitoring. That’s why we’re committed to leveraging its potential responsibly, efficiently, and securely. If you want a no-obligations demo about how we use cutting-edge generative AI tech to help lenders identify risks early and monitor their portfolios without gaps, reach out to us at info@traice.io or schedule a demo with us today! 


Comments


final1.png
Subscribe to the TRaiCE blog
Get our posts delivered straight to your inbox

Thanks for subscribing

bottom of page