Driving Intelligent Data-driven Debt Collection

Driving Intelligent Data-driven Debt Collection

business prediction and analysis insights image
Conneqt’s Unique Solution Delivers Superior Agent Performance And Enhances The Monthly Collections By 28% For A Major Motor Finance Company

A leading in-house financing company for a major automobile manufacturing company in India had considerable consumer default in their business which was impacting their bottom-line. 

They turned to Conneqt Business Solutions to optimize their end-to-end debt collection process. By deploying specially designed methodology, Conneqt helped in proactively identifying debtors who were more likely to pay with little intervention. There were two major data challenges – very few relevant variables and a highly imbalanced target variable. 

Using our intelligent approach, we helped the client overcome these challenges. As a result the client’s monthly collections rose by 28%. In addition, collection agents could focus their efforts on a more promising subset of the population, thereby, improving their performance.

Business Challenges

In an industry characterized by customer default, a leading motor finance company in India had a dire business need to optimize its final debt collection processes so as to improve the recovery of receivables


The company sought our expertise to improve their debt collection process as they had over 120,000 customers in default and was already in a loss on their accounts. The client was also not willing to allocate additional resources for the collections process. 

Prior to partnering with Conneqt, the Financing Company sorted the debtors in descending order for Amount Owed, and assigned the data randomly to one of their 30 collection agents. This method would yield $73,000 in collections a month (INR 5.4 million). 

The client also had severely limited access to the debtor’s financial information like income, savings, and other types of debt. Variables usually deemed important in any debt related predictive model like credit score and education levels were also unavailable.

Solution Components

  • As a financial and lending company’s ability to collect efficiently on its debts in today’s competitive market depends greatly on their ability to use the historical data efficiently. Therefore, Conneqt started with an exploratory analysis of historical client data, which consisted of their call records and prior payment information. The analysis enabled us to pre-empt the possible default events and predict payment propensities with accuracy.
  • We defined the target as Payment Status – a binary coded variable depicting whether or not a person has made a payment. Only 0.5% of the debtors in the historical data had previously made a payment in full. The highly imbalanced nature of the data was challenging, as predictive models could predict every debtor as not paying, and still maintain 99.5% accuracy.



At Conneqt, we follow a four-phased approach especially designed for data acquisition, data preparation, data analysis, results, and analytic models. 

Phase 1: Predictive Characteristic Analysis 

  • Combining different data sources into one and auditing 
  • Data cleaning and transformations, if necessary
  • Initial data discovery: exploratory data analysis and summary statistics of important variables 

Phase 2: Feature Selection 

  • Finding out which variables affect the outcome variable 
  • Using methods like correlation, cluster analysis, and principal component analysis we can separate out the variables to retain in the predictive model 
  • Use mathematical transformations to create new features, if necessary

Phase 3: Model Prototypes 

  • The platform will create several predictive algorithms and will rate them on performance (accuracy) 
  • Best model chosen and deployed in the database for near real-time predictions. 

Phase 4: Model Maintenance and Fine Tuning 

We schedule regular meetings with the client to discuss model performance and make changes to the model inputs if necessary. Our team continuously monitors the performance and ensures the client is using the resource efficiently.

The detailed process flow is illustrated in the figure below. 

solution 2


Conneqt’s intelligent solution and services help deliver exceptional business outcomes and customer experiences. The efficient and effective deployment of our solution resulted in staggering 28% surge in average monthly collections for the motor financing company.


Hunting down customers and asking them to make a payment is an expensive process. Indeed, the entire debt collection process can be a loss inducing event. This is particularly true if the collections are outsourced to a third-party agency. Such agencies typically have a large number of open cases. This is because it’s impossible to personally call each debtor. So, a portion of debtors are never contacted effectively. 

One way to mitigate those losses is to focus on a subset of the debtors that are deemed as likely to make a payment. If a debtor’s likelihood to pay was known, the collection agents could better focus on those promising debtors first. This would maximize agents’ time and efforts, thus potentially lowering costs of collections. 

We had to predict the likelihood of a debtor paying strictly based on previous calling behaviour and payment history. 

A confusion matrix highlighting how highly imbalanced data affects the predictive powers of a model

We used cross-validation to create an optimal sample size of the data and assigned a high cost to the error of ‘Actual=Paid|Predicted = Will Not Pay.’ This subset was used to train the model. This process was repeated a few times to ensure a random sample and model robustness.  

We identified variables that have a significant impact on the target and created new variables as well. Important variables were Amount Owed, Number of Call Attempts in the last 2 months, whether the debtor ever had Promised to Pay, and whether the debtor had previously answered a call. 

  • Negative correlation between Amount Owed and Probability to Pay.  
  • A debtor who previously has Promised to Pay is 22 times likelier to pay than others. 
  • High positive correlation between Number of Call Attempts and Probability to Pay.
  • A debtor who has previously answered a call is 73% likelier to pay than others

result 2

The output is a ranked list of debtors based on probability to pay. The collectors call the debtors from high to low probability until the list for the month is exhausted, or a new list is made.  

Benefits Delivered to the Client

Our deep business intelligence and data analytics skills and excellent customer and team management powered by Conneqt’s methodology and solution allowed us to holistically evaluate the client’s end-to-end collections process. Our client is experiencing numerous benefits post its engagement with us:

  • Substantial increase in debt collection: The increase in debt collection significantly boosted    revenues.
  • Improvement in customer experience: The reduction is average handle time lead to enhanced    customer experience and agent productivity.

Lower cost of collections: As collection agents focused on the promising debtors first, the    approach helped maximize agent’s time and efforts, thereby, potentially lowering the cost of collections.

how can we help you?

Contact us at the Consulting WP office nearest to you or submit a business inquiry online.

A global leader in Health and Fitness space based out of California, honor’s Conneqt with the prestigious “Annual Partner Excellence Award”

Congratulations from Beachbody to the team of Conneqtcorp
Annual Partner Excellence Award
, Beachbody @ California

Looking to win customer experience game?


error: Content is protected !!