Collection Scoring in the Age of Artificial Intelligence

Blog Post

Twenty-first century debt collection is backed by a complex system of inputs, algorithms and analysis involving gigantic datasets.

Almost by definition, all lenders are collectors. So, while a certain amount of delinquency may be understood as a cost of doing business, that cost doesn’t have to be met with shrugs; not anymore. Artificial intelligence, or AI, can help lenders identify and cope with borrowers who may be predisposed to come up short now and then.

To start, let’s define a few terms. In lending, the word “delinquent” refers to accounts with payments overdue but still within a pre-agreed time frame — say, 180 days, as it tends to be in the US credit-card space. “Default” comes into play when the stated period has lapsed with payment still behind schedule. Typically, accounts that are delinquent or in default — collectively “bad debts” — trigger different responses (reminders, penalties, etc.) from lenders as dictated by internal policies and operative loan agreements. 

Unmanaged, Delinquency Can Diminish a Loan Portfolio’s Value

To be sure, the benefits extending credit to consumers easily outweighs the overall impact of bad debts. This holds true for enterprises large and small, and whether or not they’re primarily financial. 

But bad debts are still — axiomatically enough — bad for business. A trickle of loans made to consumers who may be hard pressed to consistently honor repayment schedules is bad enough. That rivelet can become a flood in periods of economic stress or where, even in the best of times, systemic safeguards are lacking. Taken to extremes, this trend can put lenders in danger of total failure.

Meanwhile, consumers around the world are borrowing more than ever. Global household indebtedness went from $15 trillion in Q3 1997 to $44 trillion in Q3 2017, according to the Independent, a UK newspaper. This 193% hike isn’t entirely negative, however. As a percentage of gross domestic product, the debt burden of households around the world rose by a comparatively mild 41% in the same period. But the fact remains lenders face more borrower risk now than in the past. 

AI is a Catalyst for Debt Recovery, and a Source of Business Metrics

Fortunately, lenders can deploy AI as part of their best practices in collection strategies. In this context, it’s vital to emphasize we’re not talking about credit scoring — in which there is of course a momentous role for AI — but about collection scoring. Credit scoring helps lenders determine whether to make a loan, and on what terms. Collection scoring, our subject today, helps them figure out how to retire loans safely, efficiently, and fairly — while predetermining customers’ propensity to pay and improving recovery rates.

Collection scores can also be important to valuation metrics in several contexts. First, they can help determine the value of a debt portfolio before it’s sold to a collection agency. And of course, they can play a role in the lending company’s overall value, primarily for M&A purposes.

Accurate and effective debt-collection strategies depend on an array of variables, such as loan product, geographic region, the borrower’s behavioral profile, the stage of delinquency or default, as well as the amount owed in principal, interest, and penalties. These inputs are matched with vast statistical datasets and predictive modeling to inform AI “logic” around collection scoring. 

Not all Borrowers Think, Act or Respond Alike (Not by a Long Shot)

For maximum efficiency in this context, it makes no sense to treat all borrowers the same. After all, the vast majority of borrowers will never become delinquent. The solution? Wait for debtors to fall back on the repayments before doing any collections. Then deploy best practices around modeling and analysis of the debtor’s likely behaviors and responses to various collection tactics — and do this before other action is taken. This way, a lender knows from the very start of how easy or difficult it may be to get a newly delinquent borrower back on track.

Among other outputs, the most significant and widely used predictive models for collection scoring provide information on:

  • The amount owed and value to the lender of the loan
  • The amount due now to bring the account to compliance
  • How likely the borrower is to pay the debt in full and on time
  • How likely the borrower is to take action without external prompting to restart payments in the early stages of delinquency
  • How best to communicate with the debtor

The How and When of Communication is Key to Debt Recovery

The last point is crucial. Debtors are individuals with unique challenges keeping them from repaying, unique reasons for wanting to get back on schedule, preferred ways to communicate, and modes of outreach they’re most likely to respond to. AI helps lenders on all those fronts by delineating collection interactions based on the right:

  • Channel (post, telephone, email, text, instant messaging, lender’s customer portal)
  • Person (debtor, guarantor, parent, spouse)
  • Place (home, workplace)
  • Time of day (morning, afternoon, evening)
  • Date (time of month, special events such as birthdays, etc.)

Whether it involves identifying delinquency in its first stages or determining when and how to send messages calculated to have positive effects for lender and borrower alike, twenty-first century debt collection is backed by a complex system of inputs, algorithms and analysis involving gigantic datasets. The aim of all this computation is to evaluate borrowers, both as individuals and as loan-portfolio components, and do it quickly, accurately, and in ways most likely to result in the timely repayment of the debt in question.

See how we can help you with your collection scoring needs here!

Expand your
lending business
knowledge

Stay tuned
for more
articles