Abstract: An African lender operating in 11 countries (13 operations) developed the need to introduce credit scoring into their businesses for the first time. There were many challenges. Firstly, conducting 14 separate scoring projects (each requiring multiple scorecards) would be expensive and secondly, some books were too small to support the building of an individual bespoke scorecard. There was however a solution: The African Master Scorecard. The solution was successfully deployed within Principa’s SmartSuite and this paper tells the story.
An African lending group operating in 11 countries across 13 operations, had been offering deduction-at-source loans and small business loans and had just started to introduce individual unsecured loans. Decisioning had been judgmental and manual. This approach suppressed growth, particularly with the introduction of relatively higher risk of unsecured retail lending. It was thus decided to introduce scoring across all operations to improve the decision-making process, to ensure consistency and to encourage growth.
The challenge facing the lender was the fact that books were relatively small (particularly in countries that were just starting to introduce unsecured lending or small business lending). A particular benefit however was that in the data that they held, the lender had the collective wisdom off which they could leverage. In other words, the Lender was in a position to use data it already had to firstly understand behaviour in countries with sufficient data or across multiple countries, and could then, based upon this wisdom, infer this behaviour in similar portfolios where only limited data was available. The solution was the African Master Scorecard.
The principle underlying the African Master Scorecard (AMS) approach, was to build an AMS for each product segment, for example deduction at source (DAS), SME (MSE), unsecured loan, new, repeat. Here data from all business units are pooled together and the AMS is built. The resultant model is then calibrated on data for each country: where significant data is available – a full country calibration is conducted; where there is not – countries are clustered together and a calibration is done on that specific cluster. Ultimately, multiple scorecards were built, adjusted from the African Master Scorecard and deployed within Principa’s decision engine, DecisionSmart. The approach was also efficient in that data structures were identical and thus consistent across different countries.
This AMS approach not only saved the business an estimated 70% of the cost of conducting 13 individual scorecard projects, but also ensured predictive scorecards were deployed across each business unit. The ultimate performance of the scorecards has yet to be determined. The scores, however, are monitored monthly and rebuilds and recalibrations will be conducted when inevitably, population shifts.
Although the AMS was principally done for originations, specifically for the 3 business units (DAS, MSE, unsecured) and across the two segments (new and repeat), the same principle was rolled out for anti-attrition and collections. Since many individuals and companies took out loans, it was anticipated that many would also consider extending these loans or take out new loans. It was important for the lender to offer extended loans to good customers before such customers could approach a competitor for the purpose of taking out a new loan.
The lender chose Principa, not only for our scorecard building pedigree and our innovative African Master Scorecard concept, but also because we have the software within which to deploy the solution. The software utilised included the following (to find out more about the software, click on the relevant link):
- DecisionSmart – used to deploy the scorecards, policy rules, terms of business, existing customer campaigns.
- BridgeSmart – used to integrate with the lender systems such as the loan originations system (LOS) and multiple credit bureaus in each country.
- Stratus (Data Management Platform) – used to create aggregated data from the host system which was used for scoring existing customers. In future, it will be used to deploy machine learning models.
- DataHub – this stores all credit data and is used in decisioning flows and for reporting. Scorecard performance reporting, for example, is conducted through the datahub.
- CollectSmart – this is the collection management system that is used for the operational management of accounts in collections.
Thomas has 18 years of experience in data science focusing on the EMEA region. His experience traverses multiple industries and disciplines covering analytics, consulting and software solutions for companies ranging from large banks and retailers to telcos and manufacturing operations. A large part of his experience comes from working with credit providers helping them harness the predictive power of data through the use of machine learning and decision tech. Thomas holds an MA in Mathematics from Edinburgh University.