(By Anand Agrawal)
Machine Learning and Artificial Intelligence have proved to be the most disruptive technology so far, making its presence felt in every sector. This technology has been sector agnostic, thereby solving multiple use cases by automating most of the critical processes and reducing human intervention. It is known that the introduction of the right technology to solve/optimise the problem improves the customer experience and brings efficiency into the system. ML/AI make use of the available or generated dense data to bring out patterns to predict the next steps. This helps small size companies to leverage data and scale much faster, including such technological solutions as part of their product offering.
One such financial sector product is lending. The shift in the financial service sector in the past two decades has been mind-boggling with the use of advanced technology. Banks and financial institutions issue different types of loans—like Personal Loans, Business Loans (SME), Mortgage Loans, etc.—out of the funds collected from the saving account depositors and the money borrowed from institutional investors. The spread in the interest rates of these two stakeholders makes the Bank/Lending business viable.
The critical problem is the management of delinquent loan accounts, such loan accounts where the borrowers have missed on their payments for a duration. There are three such kinds of borrowers, one who have unknowingly missed on the payment dates, second who because of the circumstances have not been able to pay, and third, the most important, who are not willing to pay. Institutions’ asset quality gets affected tremendously because of delinquent accounts which further go on to form the Non-performing assets (NPA) once the delay in payments crosses 90 days. This rising issue of NPAs affects the lending market and the measures to keep the NPA in check become challenging for these institutions.
Traditionally, the lending institutions and banks have relied on mails and telecommunication to derive payments from customers, while managing the data over excel files. At present there are platforms which use AI and ML resulting in the prediction of the recovery chance for a delinquent borrower, based on several available data points. These points include geographical location, EMI payment history, similar set of borrowers, collection data, etc. This prediction helps the lenders adopt an appropriate strategy to recover the loan, while enhancing the customer experience. Debt collection software market size is growing at a CAGR of 9.6%, as stated in a report by Markets and Markets.
The next set of upcoming trends in debt collections optimisations with the use of technologies includes:
Early Warning Signal: Recent research has revealed that one of the primary reasons for accounts becoming delinquent is the lending institutions’ inability to identify stressed accounts that are likely to default. Several countries have rule based surveillance systems for such early warning signals over income recognition, credit quality, and trend in the utilisation of funds. An advanced early warning system based on AI using publicly accessible social activity data and news on borrowers, combined with the above financial information, will help mitigate the risk of accounts becoming delinquent. ML tools like Predictive Analytics, Natural Language Processing (NLP), and Clustering will raise alarms on such accounts. Credit restructure, credit counselling, active follow-up on such customers can be initiated by banks in such scenarios, helping to improve customer loyalty.
Intent Identification and Compliance Checker: Calls made by agents will be tracked by ML speech-to-text models to identify the misbehaviour of agents. This will help financial institutions in meeting compliance requirements. Intent identification is a second use case, where based on the audio conversation of the recovery agent and borrower, the model can help identify users with unwillingness to pay.
Reminder Automation: Use of AI/ML will help automate timelines to follow-up call and reminders, w.r.t., content and time of the day based on previous data used to train similar borrowers.
Digital Communication Journey/Recommendations: Currently lenders have been sending daily reminders with almost similar content through all the channels, without any intelligence in place. The digital communication journey can be designed for handling novelty, while overtime recommendation based customised communication channels brings efficiency in the recovery approach. For example, a borrower has opened the WhatsApp message five times out of 7 messages sent versus zero interaction over email, the platform enables the next set of communication to that borrower on WhatsApp, including sending payment links over the conversation with agents. The content of such templates would be customised, with specific content including relevant borrower information gathered at the time of trigger.
Automated Voice Bots (Robo Calling with IVR functionality): Automated voice bots will help increase the customer coverage speed with speech–to-text based models to conduct preliminary automated personalised and contextual conversations with the borrowers, thus reducing the task of manual follow-ups and are cost-effective solutions. Custom call flows can be designed based on the normal conversation trends in an agent–borrower conversation in human-like voices, in multiple languages.
Recovery Chance Predictor: ML based Recovery Chance Predictor model based on several collections, EMI payment, and geographical information helps prioritize/deprioritize the accounts for loan recovery – for example, an account with 80% probability of paying back versus an account with 20% chance. Account with higher chance would be prioritised for collections—lower cost and time but higher return on dollar spent for collections. With the accounts on the lower side of recovery chance (generally on higher dpd), legal approaches could be opted for such borrowers.
Summing it up, over the last few years the NPA rate has shot up across countries, demanding the urgent need to reconsider existing debt recovery practices. Globally, banks need to switch from an increasingly compliance-driven post-facto process to a proactive digital framework focused on controls to track credit risk. Debt collection has been the critical part of any lending institution and advanced technologies like AI and ML-powered models that can help track real time identification of stressed accounts, aid in resolution, and automate communication flows, thereby reducing bad debts. As the data continues to get densely populate, more advanced models will keep evolving in this sector to optimize the debt collection processes.
(The author is the Co-Founder and CTO, Credgenics)
