The Data Analytics Blog

Our news and views relating to Data Analytics, Big Data, Machine Learning, and the world of Credit.

All Posts

Incorporating Credit Lifecycle Predictive Outcomes In Your Collections And Recoveries Call Centre

January 16, 2019 at 10:07 AM

In a collections environment, an agent needs to follow up with numerous customers on their outstanding credit and the more distinct information the agent has on each customer, the better the agent will understand who they are interacting with and what the opportunities, risks and expectation of the collections call with the client are.

We’ve identified four models, where the outputs could assist collections agents in reducing bad debt, improving collections and reducing needless follow-up.

Models to use in your collections call centre

Propensity to pay (PTP) model, given right party contact (RPC)

A PTP model, given RPC, allows an agent to understand the likelihood of a customer making a payment when the number they have dialled reaches the correct person. By recognising how likely a customer is to make a payment, an agent can decide on how to treat the customer. For example, if a customer has a high propensity to pay, the agent may feel a quick reminder is enough to ensure a payment. However, if a customer has a low propensity to pay, the agent may want to spend more time explaining the repercussions of non-payment; discuss ways of how to make a payment; etc. to better ensure a payment is made. Similarly, if a client has a high propensity to pay but during the call with the client, the client is hesitant to commit to a promise to pay this insight could inform the agent to invest a little more time and effort to attain a promise to pay due to the higher likelihood that the time is well spent and would lead to a promise.

Propensity to default (PTD) model, given a promise to pay

A PTD model, given a promise to pay, allows an agent to understand the likelihood of a customer breaking their promise to pay. By having this information on hand, the agent can adjust their conversation with the customer accordingly. For example, if a customer has a low propensity to default, the agent may feel a quick confirmation of the promise to pay is enough to ensure the promise is kept. However, if a customer has a high propensity to default, the agent may want to spend more time explaining the repercussions of non-payment; confirm the source of payment, ensure the level and timing of the promise aligns to the customer income and discuss ways of how to make a payment to better ensure that the promise is sincere and that the promise is kept.

Preferred installment model

A preferred installment model aims to predict the largest amount a customer is likely able to pay. This amount could be used by an agent to better negotiate the promise to pay value. For example, suppose the preferred instalment amount is R200. If the client promises to pay R100, the agent could try to persuade the client to pay more, knowing the client can likely afford to increase the payment amount. However, if the client promises to pay R300, it is possible the client cannot afford to make the payment therefore increasing the likelihood of breaking the promise to pay. Given this, the agent could suggest to the client that he rather commits to a lower payment and if the opportunity exists, to make further payments later on.

Settlement amount model

A settlement amount model gives an agent the best account level settlement offer that should be negotiated with a client to maximise the lifetime collection return of the portfolio.  The account level settlement offer is analytically defined by evaluating each client’s expected lifetime recoveries to ensure that expected payments from expected paying customers are retained while increasing payments from expected non-paying clients.

Account level settlement strategies mitigates the risk of offering the client a settlement discount and losing out on the present value of lifetime expected cash-flows / payment. For example, taking an immediate hit on the impairment, being bespoke to the customer needs, introducing tiered offers for negotiation and considering the behaviour of expected paying accounts only.

Having the account level settlement offers displayed to the collections agent makes the option and negotiation of a settlement offer top of mind.

Model methodology

The first two models use a very similar methodology to predict a likelihood of payment following an RPC or a likelihood of default on a promise to pay. The typical definition and outcome period of a payment is R50 in the month following RPC while that of a default is as per the broken promise flag in the collections system. The models will use past behaviour and trends to predict future behaviour. Therefore, an extensive list of features based on historical data will be created to use in the model development. Once all of the above has been established, predictive modelling techniques will be used to model the outcome. These models rely on there being sufficient historical behavioural data for modelling and feature creation.

The preferred instalment model will predict the probability of default for a range of promise to pay values. The typical outcome period of this payment is in line with the promise to pay due date while that of a default is as per the broken promise flag in the collections system. After creating an extensive list of features based on historical data, action-effect modelling will be used to determine the probability of default with each promise to pay amount. From this, either the maximum expected value for each account holder can be calculated (i.e. expected payment value, given the probability of default), and/or a maximum default threshold can be applied (i.e. no values above a certain default level will be considered).

The preferred instalment model relies on there being sufficient data in terms of the range of promise to pay values with some account-holders breaking their promise to pay while others keeping their promise to pay.

The settlement offer model will predict the expected discounted lifetime recoveries over current outstanding balance for each account. The predictive model will use a combination of short term and long term past behaviour and looks to predict future payment behaviour. Short term data is important to ensure that the models remain accurate to the most recent recovery strategies and performance and long term data indicates the longer term shape of payments. Statistical/empirical modelling techniques will be applied to predict the payment over balance ratio in the short term and then these will be aligned through scorecard alignment processes to a lifetime discounted payment expectation for each account.

The modelled lifetime recoveries rates will then be statistically adjusted to determine a maximum settlement offer percentage that will ensure that the current expected payment levels from paying customers are retained while increasing payments from expected non-paying clients, i.e.:

  • If all customers that are expected to make at least one payment (regardless of a settlement offer) take the maximum settlement offer the net present value of the total payments of the settlement campaign and that of expected payments of the lifetime of the account will be the same.
  • For every non-paying client taking up a settlement offer, the payments received equals a benefit to your organisation.

The modelled settlement offer percentage for each account and a tiered settlement offer strategy, allowing for different levels of negotiation and multiple payment options, will then be implemented and modelled percentages will be converted into 3 tiered Rand value discount offers for each account.

Data required for these models

The data requirements for each of these models are largely similar. The following historic data will be required:

  • Account performance data, payment information; account age; delinquency; amount outstanding; etc.
  • Account-holder data, including account-holder demographics (e.g. age, gender, etc.)
  • Bureau data for SACRRA members

To develop a PTP model, given RPC it is additionally required to have historic data for:

  • RPC flag with date stamp

To develop a PTD model, given a promise to pay and the preferred instalment amount model it is additionally required to have historic data for:

  • Promise to pay level data over time with account link, including amount, terms and break or kept indicator

Information displayed to your agents

It is vitally important, however, that you call centre agents have easy-access to the outcomes and recommendations made by these models.

With models built by Principa in our proprietary decisioning software, DecisionSmart, Principa’s Agent X Call Centre Virtual Assistant can deliver real-time call and account specific data, decisions and information to your agents visually, to inform, guide and motivate them.

Read more on Agent X or request a demo.

Contact Us to Discuss Your data analytics Business Requirements

Megan Wilks
Megan Wilks
Megan Wilks is a consultant within the Decision Analytics team at Principa. Prior to joining in 2016, she worked in the consulting and data analytics field, covering a range of industries, including retail, credit and non-profit in Southern and East Africa.

Latest Posts

Solving the Credit Unaware Challenge with Psychometrics

At Principa, we engage with clients and organisations across the entire credit lifecycle and track the focus of the South African credit industry. For nearly ten years the focus has consistently been in the collection space, but recently (since early 2021) this has changed and a large number of our clients are focused on acquisitions and originations.

Predicting Customer Behaviour (PART 2)

In Part One of this two-part blog, we started providing a short overview of just some of the propensity models that Principa has developed. In this Part Two, we continue to look at different types of propensity models available across the customer engagement lifecycle that are used to predict behaviour and solve business problems. 

PART 2: How to Cure the Post Pandemic “Collections” Symptoms

In PART 1 of this two-part series, we explored how the current socio-economic climate resulting from the lingering financial hangover caused by the pandemic is negatively impacting the consumer's ability to settle a debt.