Collections efficiency: The playbook to combine the best of AI and human intelligence

Aparna Chandrashekar   /    Content Specialist    /    2022-01-21

LinkedInLinkedIn

To say that current debt collection practices have an image problem is to understate the negativity with which they’re associated. Threatening, incessant phone calls, being strong-armed for money, or in more severe cases - fraud, hounding and social shaming. It’s all part of the package if one defaults on a loan from a lender. 

A quick look at the Reserve Bank of India (RBI)’s concerns with debt collection agencies from 2008 and 2021 and you’ll realise not much has changed - uncivilized, unlawful and questionable behaviour and recovery processes continue to rule the roost. Except now, it’s only getting worse. Add a layer or 100, of additional information that lenders (and debt collection agencies)  are privy to and we have ourselves a data privacy problem to deal with as well. 

The RBI believes that the ultimate onus of borrower safety, both physical and digital, is on the regulated entity. In the recent past, there have been several reports of recovery agents using borrowers’ phone contacts, photos or any other sensitive data to harass not just the borrowers but also their friends and family as a form of extortion. 

But how did we get here? In the absence of an on-field debt collections team, some lenders misuse signed agreements to access mobile phone data and contacts of borrowers to threaten borrowers into repayment. 

This absence of on-field collections teams and a complete move to digital collections has also resulted in limited segmentation - most collection models are based on limited data points that don’t reflect changing economic conditions. These models can’t predict which loan accounts are likely to default or of the defaulted ones, which can still be salvaged. 

The Case for a hybrid collections model 

The core of an orderly, intelligent collection model is to proactively identify the vulnerable. It’s worth mentioning that collection departments are now tasked with dealing with borrowers who aren’t defaulting for the usual reasons - they may have been furloughed with a rapid, significant decline in income and these sudden shifts in circumstances have yet to be detected by credit bureaus. 

India’s household debt to GDP ratio rose to 37.3% in 2020-21 from 32.5% in 2019-20. This percentage is expected to rise further in 2021-’22 due to depleting bank deposits as families carry the burden of medical expenses incurred during the second wave of Covid-19. 

Most lenders are proactive about pre-delinquency; however, there’s a real need to dig deep across the organization to capture data and find information that's vital to demarcate economic victims from the steady state collections customers. 

The RBI in its most recent report on digital lending recommends a hybrid model of collections. Having an on-field collections team and an optimum-sized call centre would make the lender understand the challenges faced by the customers in repaying. And it makes sense.

With intra-day, reactive changes in a COVID world, there are more people on arrears that were not expected to be there, and the data you had four weeks ago is now old - it’s critical to understand which customers are impacted by COVID, for instance, and likely to bounce back and which would be in financial trouble irrespective of the crisis.

Although this is resource-intensive, it’s also one of the only ways to get data points like

  • Reasons for repayment failure: Unemployed, furlough/reduced income, medical, quarantined.

  • Type of relief applied: Payment holiday, reduced instalment, repayment duration changed – what relief did the customer have and when did it stop.

Leveraging hybrid data

The legacy of Artificial Intelligence(AI) and Machine Learning (ML) models have conditioned us to believe that they operate without human input. For instance, when Google defended its use of human resources to improve its understanding of voice conversations for Google Assistant, many were shocked. 

AI & ML models efficiently cull data, in record speed, and reveal insights about delinquency risk and how to manage at-risk accounts. 

Collections are not about reminding customers to repay overdue installments alone, it goes beyond that to suggesting a way out of an imminent crisis and this is where AI comes into play - workflow automation and borrower segmentation will drive process efficiencies and free up resources for more value-added services. 

New age debt collection models then include early warning for delinquency, better borrower categorisation, and optimised strategies for customer engagement to reduce defaults.

Early warning system

Debt collection has historically been reactive; post delinquency loss recuperation was the norm. ML changes this paradigm. It helps identify borrowers that will potentially default by mining insights that were previously unidentified. Finbox’s CollectX, for instance, identifies risky borrowers by collating data points such as low balance, multiple credit obligations, upcoming loan dues, increased credit utilisation, low financial inclusion score. This identification acts as an early warning system for lenders. FinBox CollectX flags risky borrowers five days prior to their due dates, so that lenders can allocate collections resources strategically and use targeted communication.

Borrower categorisation 

Lenders can build a nuanced customer profile to recognise which borrowers are likely to resolve delinquencies on their own and which borrowers need intervention (loan restructuring, modified repayment terms etc). Real-time cash flow data, like account balance, credit appetite, over indebtedness, etc gives lenders precise insight on the borrower’s financial position post disbursal. It generates prioritisation buckets against each borrower, allowing for tailored communication, as opposed to the one-glove approach that leads to an unquantifiable drop in loyalty. At Finbox, we’ve noticed that bucketing borrowers according to risk helps lenders lower cost of collections by prioritising resources. Additionally, a smooth collection process improves customer satisfaction. This means they are more likely to convert to reapply for credit from the lender. 

Optimising engagement 

There are multiple platforms to reach out to your borrower - emails, text messages, social media, mobile apps, website chat bots. But the problem is not the availability of platforms, it’s appropriately choosing which platform, knowing when to reach out and crafting an effective message. 

What most lenders fail to recognise is that these elements of engagement are all contextual and depend on various factors. For example, lenders can design a customized outreach strategy based on their mobile app, where they can identify a preferred method and time of engagement that can be integrated with demographic/financial information. Audio from customer calls can also be used to determine how different scripts and offers impact customer response and collections thereof.

Conclusion 

AI has always stood on the shoulders of human intelligence. It may be far superior in its ability to predict delinquent behavior, but it still has much to learn from our uniquely imperfect perspectives. A deeper understanding of your customers and the ability to identify and interact with borrowers intelligently is ultimately a mix of doing the dirty ground work and AI. 

In terms of collections, steering away from a reactive approach to a more proactive customer outreach can stem delinquencies, along with borrower penalties, credit markdowns and potential insolvency. 

Finbox’s CollectX is built to be proactive - it seamlessly integrates with your existing decisioning systems and helps lower cost of collections, delinquency rates, and improves collection efficiency and eNACH success rate by recommending most-suitable dates, and increases the number of repeat customers.

You can go live with CollectX in 3 days or less. Let’s connect here.