For banking majors, credit risk has always been a challenging area, given the multiple factors that go into forming an individual’s risk profile. For business borrowers, the process is even more complicated as data across a variety of parameters and time periods must be aggregated and analyzed to create a holistic picture of risk.
And the stakes are extremely high for lending banks -- inaccurate assessments can cost organizations sizeable amounts. This is further intensified by sub-optimal underwriting, inaccurate portfolio monitoring methodologies, and inefficient collection models.
Clearly, it is imperative for banks to adopt smarter models of credit assessment that can parse huge volumes of data in truncated timelines, dynamically altering risk profiles as per real-time data. This is where Artificial Intelligence (AI) and Machine Learning (ML) could play a major role.
These technologies are able to learn from complex datasets and become incrementally more accurate over time. Further, the need for human data science expertise and analysts’ efforts is also minimized, as AI & ML models can “black box” the underlying technology to show only the final insights.
The State of AI & ML Application in Credit Risk Assessment
AI & ML technology could find a plethora of use cases in the BFSI sector, and risk management is at the top of this list. Between 2017-18, the number of organizations using AI more than doubled, and 40% of financial services firms are applying it to risk. This is because AI & ML could add genuine value across the credit value chain, starting from the initial underwriting process to risk measurement and analysis, until deciding on the final maximum exposure. Some of the key use cases that would be addressed are:
- Assessing risk for individual customers - banks typically deal with a diverse pool of borrowers, each with a unique risk profile and exposure parameters.
For example, the majority of applicants in any borrower group will comprise non-defaulting individuals; only a very small set will be defaulters. As a result, a sample of bad customers going into the credit dataset could cause imbalance and skew results if banks are using traditional analytical models. Because of performance degradation caused by this gap, the predictive insights are inaccurate, and the bank loses out on viable business opportunities.
An ML model, like the Artificial Neural Network, would create discrete clusters of datasets and apply merging methodologies to figure out if a specific customer should be offered a loan. Instead of merely looking at the mean values, ML creates majority and minority clusters and merges them to create a diverse dataset, reflecting the real on-ground picture.
- Rating the credit score for a commercial entity - Assigning a credit rating to any company involves sizeable investments, time, and expertise. Typically, banks would employ specialized credit rating agencies to conduct a rigorous evaluation based on financial as well as non-financial indicatives.
Unfortunately, it isn’t always possible to invest at this scale for companies that are relatively young or operating in a nascent industry sector. Consequently, the company is denied the capital required to propel growth, and the bank, once again, misses out on an opportunity. AI & ML turns this on its head by capturing the various transaction patterns that make up a company’s risk profiles.
Parameters such as market resilience, management changes, currency flux, ethical stance, etc. that are typically viewed as “subjective” could be analyzed using AI to make robust risk modeling more accessible.
So, what are the technologies that could fuel this transformation in the banking sector? Broadly, it can be classified as Artificial Neural Networks, Random Forest, and Boosting ML techniques.
Understanding the ML Techniques that Could Reshape Credit Risk Analysis
AI & ML offer a significant leg-up over traditional statistical models. Instead of defining a set of rigid instructions to arrive at a particular insight, ML can adapt and intuitively “learn.” The ML model is continually fed data, made to extract insights, and then draws predictive insights on new datasets. This is a cyclical process, meaning the ML gets more accurate with every round of credit analysis. Here are three ways in which data could be processed and understood using ML:
- Artificial Neural Networks - This is a mathematical representation of a biological neural network that is equipped to manage non-linear and interactive patterns between variables. In cases such as company credit scoring, Artificial Neural Networks could prove game-changing.
- Random Forest - This creates decision trees with each tree dependent on a random sample set with an equitable distribution of data. Insights gathered from multiple trees are aggregated, and the average denotes the final outcome. For areas such as personal risk analysis, where diverse datasets could skew results, this ML technique is particularly relevant.
- Boosting - This is a version of Random Forest where each underlying tree is valued as per their relationship with the cumulative dataset. Instead of considering the average, the results of one decision tree are taken into account before processing insights from the next. Expectedly, Boosting could offer more optimized results than the Random Forest approach.
Clearly, the AI & ML methodologies needed to modernize credit risk assessment are already in place. And with the advent of Big Data technologies that makes unstructured information part of the risk assessment process, these models are poised to be more accurate than ever before. So, what is holding back industry-wide adoption?
Exploring Adoption Opportunities: Early Experiments and The Way Forward
The black-boxed nature of AI & ML could be a factor holding back implementation, especially among larger regulated entities. There’s also a risk of overfitting the data, as ML models are more sensitive to outliers than traditional analytics. However, these could be termed as “teething troubles.” As AI & ML continue to evolve and innovators find that critical balance between transparency and performance, the true potential of this technology will take center stage. Here are four examples that drive home the incredible potential of AI & ML:
- FinTech startup ZestFinance has used ML to optimize credit risk for banks, reducing losses and default rates by 20% on average. Similarly, JPMorgan Chase introduced a contract intelligence platform in 2017 that used ML to review 12,000 credit agreements in seconds -- in human effort units, it would take approximately 360,000 hours every year to go through this volume.
- In a recent survey, the Bank of England found that ML is now pervasive across front office to back office processes. Credit risk management is a popular area of application, in addition to pricing and underwriting for general insurance policies. In fact, existing risk management frameworks might even have to evolve to meet the level of sophistication displayed by ML said the Bank of England, which is joining hands with the FCA to form a public-private group for innovation in this area.
- BlackRock has already undertaken the implementation of ML to accurately assess liquidity risk. Asset managers of the company will feed internal trade data into the current liquidity model, and apply an ML overlay. This will help to calculate the cost liquidating fund positions, primarily because this is a non-linear area of research, best suited to neural networks.
- Going beyond the BFSI sector, automotive giant, Ford, is actively embedding AI and Ml capabilities into its car fleets. The company has invested over $1 billion in Argo AI, a company whose innovations will find place in car manufacturing lines across the world. Interestingly, the data generated by these AI connected systems can feed directly into ML models to assess driver risk, vehicle condition, and possible insurance claims scenarios.
And this is just the beginning -- as AI & ML start to transform sectors around the world, it’s only a matter of time before every major bank sits up and takes notice. By leveraging the power of these bleeding-edge technologies, banks will soon be able to reinforce their risk posture and increase their appetite while continually exploring new levels of profitability.