Machine Learning in Debt Collection Management Process

0 Comments

Machine Learning in Debt Collection Management process

Introduction

Digital transformation in Debt Collection Management is also to use algorithms that learn to find patterns and solutions in daily information to save costs and maximize contactability.

Making it available in a Collection Management system will help to efficiently manage it.
The debt collection process is based on the use of limited resources: systems, massive contact tools (Mail, SMS) or through operators who make calls.
The recovery curve is linear, I only charge more by adding resources, but in any case it is a decreasing asymptotic function.
Since there are clients who, no matter how much they contact in a massive way or by personal management, will not be able to recover them.
If we imagine a Collection Contact Center with 15 operators, who make a total of 2,000 calls a day, we could increase their efficiency.

Not only by carrying out a good contact strategy, based on a methodology like the one below.

Processing of debt information.
Portfolio segmentation.
Definition of the contact strategy
Contact customers with operators and channels
Analysis and reports

But it will also be necessary to enhance it through an element that allows the calls to be made by discriminating the debtors based on a threshold determined by their “Probability of Payment”.

Modeling

This is nothing else than using the resources (operators, Mass Actions, System and Strategy) much more efficiently.

This is achieved through the use of a “Debt Collection Scoring” that provides a quantitative assessment of the recovery potential of a debtor.
That “adjusts” or “orders” the cases to be contacted based on that probability.
Defines the probability of Debt Recovery and the section of the management in which it is most likely to recover.
Allows generating a consistent model with good predictive power.

The Debt Collection Scoring Process takes measurable variables that it collects from the history of delinquent clients and executes with a certain periodic frequency a statistical model of correlation of said variables.

Variables

The model calculates the probability of payment associated with each value of the variables. These variables are related to the profile of the financial operation (term, amount, guarantee, type of product) to the profile of the client (sociodemographic, economic, work or retired) to its payment behavior (history of arrears, defaults, anticuación) and other variables (economic macro, geographical area, etc.)

Example

For example, that for a certain province, product or type of client there is a greater probability of recovery for a certain term. The statistical model “ties the correlations” between all the variables and calculates the specific weights of each one in the model.

Therefore, each financial product or term within the portfolio will have a different Scoring.

Conclusion

Using this Scoring and adjusting the day-to-day contact strategy, is how we can optimize the management based on a threshold that allows the efficient use of resources, not calling debtors that according to my model should pay in a way voluntary without intervention of the operators.

The Digital transformation in Debt Collection Management is a reality, using Machine Learning in the daily execution on the Debt Collection Management Strategy is an urgent need to reduce costs and increase efficiency. Machine Learning in Debt Collection Management Process is a fundamental piece to use to reduce time, cost and resources.