In a banking and financial institution, Collection analytics helps in understanding customer preferences and behaviour patterns in order to help them reduce delinquencies and mitigate losses allowing businesses to maximize their accounts receivable recovery.
It gives valuable information about the customer which can help develop varied collection strategies, to primarily improve productivity as it is not always feasible to hire different agencies (which is basically cost incurred) to contact different customers reminding them to clear their payments.
TransOrg’s predictive analytics-led collection strategy enables clients to determine which customers have a higher probability of loss in future, categorize the different types of customers as per their risk, and prioritize and efficiently target customers.
Conventionally, the banks use to segment customers based on their delinquency buckets. These buckets are basically the simplest form of customer segmentation depending upon the days since the customer has not cleared his outstanding dues after the due date (known as DPD – days past due). In the early stage of consumer default (lower DPDs), there is a higher chance of self-cure (i.e., customers are likely to pay by themselves without the need to make collection calls). And as the duration of default increases (higher DPDs) the collections effort also increases. Below figure shows the different stages of customer default –
It means that a customer who has just recently missed his payment (i.e. in low DPD) is a low-risk customer as compared to someone who has not paid his dues for a long time (i.e. 3+ months). Visualizing the evolution of late payments across different delinquency buckets can indicate portfolio health.
Now with the help of the different types of data available such as –
and integrating them by using advanced analytics as well as applying machine-learning algorithms, the banks are able to progress to a more in-depth and better understanding of their risky customers, which can be further classified into micro-segments where effective strategies can be designed for them. We can decide whether the customer is more or less likely to default in future based on the predicted output from the ML model.
Here is an example of customer segmentation basis their model risk score -
The customers have been divided into 3 risk segments on the basis of their probability of default as predicted by the model. The Red category or the High-risk segment is the one where the default rate is maximum and these types of customers are more likely to default in future, followed by Amber category or the Medium-risk segment. The Green category or the Low-risk segment has the lowest default rates, and these types of customers are least likely to default in future.
Below is another example of customer segmentation using model risk score along with another model a customer behaviour scorecard or simply a variable like total outstanding balance of the customer -
Here the output of two different models is used to create a more robust 5-tiered customer segmentation. The customers who are in high-risk category as per both the models are super risky type of customers and are tagged in Very High-risk category. Similarly, the customers who are in low-risk category as per both the models are probably self-cure type customers and are tagged in Very Low-risk category.
Collection strategies can now be designed to prioritize and efficiently target each customer segment independently. Instead of equally approaching and targeting all customers, the banks and collection agencies can now focus more on extremely high-risk customers and prevent them from going default. Alternatively, if the objective is to recover as much money as possible with keeping in mind the collection costs, then the extremely low risk segments will be aimed for because of their low default rates or high recovery rates.
Reach us to know more and build your business opportunities with AI Capabilities.
Artificial Intelligence | Nov 30, 2018»
Artificial Intelligence | Nov 30, 2018»
Your request has been submitted successfully.