How Banks Can Leverage AI to Better Manage Pandemic Relief Programs

By Rajendra Gangavarapu

On June 8, 2020, the National Bureau of Economic Research (NBER) announced that the U.S. is officially in recession. This ended the expansion of the economic cycle that began in June 2009 after the last financial crisis. The sudden contraction due to COVID-19 has caused massive economic shifts, changes in customer behavior, and payment trends. Spending slowed markedly, as uncertainty about consumer sentiment and confidence shot up.

The percentage of accounts entering “financial hardship” status has risen dramatically for all credit products (auto loans, credit cards, mortgages, personal loans, etc.) over the past couple of months. According to a Mortgage Bankers Association survey, the share of loans in forbearance grew from 0.25% as of March 8 to 8.55% as of June 7 (an 830 basis-point increase). Since these forbearance and deferment programs offered by financial institutions can provide temporary relief to consumers, there has been a significant surge in enrollment requests.

To maintain their financial stability, financial institutions need to accurately predict forbearance enrollment volume and track the performance.

Banks can leverage AI to continuously sense the macro-economic environment, to explore opportunities for loss mitigation, predict enrollment trends, and improve collection efficiency proactively in this rapidly changing environment due to the health crisis. Continuous monitoring of changes in the customer’s outstanding balance, line utilization (especially in hard hit sectors like entertainment, travel, oil and gas, etc.), and employment type; pandemic impacted locations; and ripple effects would enable timely identification of early-stage delinquencies and improve the cure rate.

Banks must stay close to their customers through digital channels and continue to enhance the relationship through effective credit counseling. They need to provide advice on the right relief program that fits each customer and must continuously track the performance of the applications and relief programs. Banks need to have a better understanding of how loans that have been deferred will be repaid and recommend the next best action to improve the cure rate and bring accounts back to current (zero delinquent state). Through timely identification of risk, banks can better manage profit and loss by proactively identifying which of the actions are optimal to bring accounts in forbearance back to current status.

Conversational AI

Banks are witnessing an unprecedented increase in call volumes and wait times. Banks must accelerate digital capabilities to ensure that their digital channels outperform the competition to succeed in this new environment. Banks can leverage natural language processing (NLP) and conversational AI capabilities to address the increase in call volume, understand customer intent and preferences efficiently, and provide a timely response. Long short-term memory (LSTM) and sequence-to-sequence learning models help in understanding customer intent, preferences, and behavioral shifts by analyzing the questions asked. Bidirectional Encoder Representations from Transformers (BERT), convolutional neural networks (CNN) and multilayer perception (MLP) can be leveraged to predict a next set of questions using customer context. This would improve the customer experience, engagement, and retention.

Business Context Graph for Leveraging Disparate Data Effectively

A business context graph can significantly accelerate data utilization of new and existing data sources (internal, external) and generate actionable insights at scale. The organization will be able to combine enterprise-wide data that exists in silos and generates contextual insights using AI. Probabilistic graphical models (PGM) such as Bayesian network optimization and Markov flow help to capture complex multi-level relationships and strength of relationships over time in a rapidly changing environment. They can provide insights on what is the appropriate forbearance program (payment deferrals, fee waivers, duration etc.) depending on the context of the borrower, track the performance dynamically, and provide recommendations for the next best actions. 

Synthetic data can be created using advanced machine-learning techniques, such as generative adversarial networks (GANs), to train new analytical models when the historical data are of little use. Reinforcement learning can be leveraged to sense the environment and make intelligent decisions, particularly learning sequential decision-making tasks.

Model Monitoring

In these unprecedented times, model decay can be faster. Hence, models must be re-evaluated to ensure that they are effective and working as expected, with more frequent model monitoring than ever before. Organizations need to develop next-generation models that leverage the recent data during the pandemic, as well as modeling techniques better suited for the fast-changing environment.

For optimal operational performance in the current economic environment, Banks should:

  • Continuously monitor enrollment volume, identify early-stage delinquency trends, and optimize collections cost leveraging PGM and business context graphs.
  • Strengthen digital capabilities using NLP and conversational AI to provide better customer experience and engagement.
  • Develop what-if scenario/stress testing capabilities to understand the potential impact of changes in the macro-economy on capital requirements, cure rate, reserves, recovery rates, and profit/loss.
  • Models need to be explainable, unbiased, and undergo rigorous processes to mitigate the risk.
  • Automate the insight generation process at the micro-segment level to improve operational efficiency, scale, and speed to the market.
  • Monitor the model performance regularly when ground truth is available to ensure models are effective.
  • As always, keep humans in the loop to understand the context.

AI can be leveraged for forbearance, payment deferrals, and collections efficiency not only in benign economic environment but also in uncertain times like these. 


Rajendra Gangavarapu is Head of Data Science at diwo. He has two decades of leadership experience in risk management, helping Banks to solve complex business problems by leveraging data and analytics. He is a speaker at various academic and industry conferences on data science, risk, Artificial Intelligence (AI) and Machine Learning (ML). He can be reached on LinkedIn.