Optimization of collections using analytics

Collection, a critical function for every NBFC, was identified as a problem area in the Client’s processes. Around 10% to clients dishonored monthly EMI payments, leading the collection process to become highly expensive without significant results. They approached us to reduce their cost of collections.

• High representation bounce rate
  Each payee with a bounced payment was given second chance to make the payment, a process technically termed representation. 90% of these attempts defaulted again, leading to heavy wastage of representation cost and efforts.
• High cost of collection
  Collecting default payments is a resource intensive job, which made it costly. Moreover, the field collection team couldn’t carry out the collection functions during the representation period, i.e. the first half of the month, bringing down the productivity significantly.
  Nexdigm was mandated to explore opportunities to address these challenges through process optimization using data-driven analytics.

Through our initial assessment and discussions with the business teams, we identified three major focus areas- effective customer engagement, improved field team productivity, and optimization of the collection process. After a detailed analysis, we converted these focus areas into two specific problem statements:

1. Representation Process Optimization
2. Effective Customer Engagement

Representation Process Optimization

• The existing representation process
  • Was spread over 14 days, impacting the productivity of the collection team.
  • Needed a lot of manual data consolidation with the involvement of multiple teams
  • Automated filtration of cases was missing, which led to low realization rates and high cost of representation.
• Solution designed
  • Developed Machine Learning (ML) backed predictive models for more filtered representation.
  • 70% of 10% dishonored payments were filtered out based on the above model and directly forwarded to the next step, saving time and representation cost. This also reduced the idle time of the field collection team and increased their productivity to almost double.

Effective Customer Engagement

• Existing process
  • The customer database had inconsistent and incorrect details limiting contact with them. The defaulters who could have been engaged with and reached out to before the EMI due dates.
  • SMS campaigns did not impact the outcome because of carpet bombing marketing approach and incorrect data.
• Solution designed
  • Developed a communication strategy backed by Predictive Behavior Models to intervene before the due date. This avoided the unintended defaults in payments and controlled future delinquencies.
  • Mapped communication channels (voice blast, tele-calling and SMS) basis the customer profile. This led to a reduction in the burden on the field collection team since a number of clients responded to the first round of communication itself.
  • Implemented database improvement initiatives.
  • Identified patterns of response by customers to multiple mode and tone of communication. This impacted customer behavior positively and brought down unintended defaults

We built machine learning models in sequence, to predict the cases to be filtered via the representation process vis-à-vis the cases to be closed through reminders by using 15 years of data and nearly 5 billion data points.

Based on our model, only 30% of the default cases were sent to representation, thereby reducing the cost of representation by around 70%. We also managed to reduce the turn-around time of the representation process by 21%, through process optimization.