Modeller Case Study with Datacuity
It is always interesting to hear directly from our partners and clients how they are using our software solutions and the benefits that they gain. This case study explores how Datacuity, an analytics consultancy based in South Africa, leveraged Paragon Business Solutions' Modeller software to develop a robust Bureau Application Scorecard for a client in Sub-Saharan Africa. We recently met with Cobus Bekker from Datacuity to gain insights into their process and the benefits experienced using Modeller for this project.
What was the main objective of developing the Bureau Application Scorecard?
The primary objective was to determine distinct population segments, create a predictive model for each distinct segment that accurately assesses the creditworthiness of applicants by using a combination of credit bureau and demographic information captured at the point of application. This scorecard was needed to help make informed lending decisions for client disbursing loans in the agricultural and informal business industries.
Why did Datacuity choose Modeller for this project?
We selected Modeller because of its powerful yet user-friendly platform that allows for the efficient development and refinement of predictive models. Modeller’s advanced features, including Weight of Evidence (WoE) grouping and visualisation, reject inference, and performance evaluation, were crucial for our project needs. Modeller contains many valuable built-in components specifically created for credit scorecard modelling compared to generic or competing software (hold-out, weights, grouping of special cases, auto handling special cases in model builds, inference reports, standard reports etc). This makes the whole modelling experience quicker, less error prone and easier to understand. The full transparency that Modeller provides is also very important to us and for our clients. Nothing is black box. The complete model development process is transparent, repeatable and auditable.
How did Modeller help in the feature selection and engineering process?
Modeller provides comprehensive tools for feature selection and engineering. We used its field reducer module to remove characteristics that showed no predictive power or were highly correlated with others. This ensured that the model development stage was focussed on the most relevant and powerful data fields. We also used the automated WoE grouping functionality to reach a very good starting point for the identification and understanding of the predictive patterns for all candidate scorecard characteristics. This streamlined the process significantly, saving time and improving our data understanding and model quality.
What were the key benefits of using Modeller during the model building phase?
During the model building phase, Modeller’s intuitive interface and robust algorithms enabled us to experiment with different modeling techniques quickly. Having a wide selection of fully supported and maintained modelling algorithms available is really valuable to us. This means we can easily investigate and compare traditional statistical techniques to modern machine learning including logistic regression, CHAID, CART, Random Forest and XGBoost approaches. We could compare various models, tweak parameters, and compare and validate performance efficiently.
How did Modeller aid in the validation and deployment of the Bureau Application Scorecard?
Modeller facilitated thorough validation of our model through out-of-sample validation, ensuring that the scorecard was not only accurate but also robust. For deployment, Modeller’s model export capabilities allowed us to seamlessly generate the scorecard code to be consumed by our client's existing scoring systems, ensuring a smooth, and easy to test, implementation.
What were the outcomes of implementing the Bureau Application Scorecard developed with Modeller?
The implementation of the Bureau Application Scorecard resulted in a more efficient and accurate credit assessment process. Our client experienced a reduction in default rates and an increase in the approval rate of creditworthy applicants. The scorecard also enabled more consistent and transparent decision-making, enhancing overall trust in the lending process.
Can you highlight any specific features of Modeller that were particularly beneficial?
We have found Modeller very easy to adopt and quickly become productive. Some standout features of Modeller which we really valued during this project included;
- Multiple model versions were quickly and easily developed. Every model version was automatically stored and we could easily retrieve or revert to a previous version. This really helped with our iterative approach to model development.
- Reporting and report templates for model performance, model validation, model comparison and model explainability. The reports are all customizable, which enabled us to easily build custom reports in line with the client’s requirements.
- Ease of model documentation. The ability to easily export reports, characteristic information and specific data points made the model documentation process a breeze compared to other software in the market.
Would Datacuity recommend Modeller to other analytics consultancies or financial institutions?
Absolutely. Modeller has proven to be an indispensable tool in our toolkit, allowing us to deliver high-quality, data-driven solutions efficiently. We highly recommend it to other consultancies and financial institutions looking to enhance their credit risk modelling capabilities.
If you would like to learn more about Modeller please get in touch here, or by emailing us at info@credit-scoring.co.uk.