Limiting credit risk, maximizing revenue
Unique credit scoring model lowers Unigro credit risk by 20%
Furniture, electronics, household appliances, towels, et cetera: you can buy almost anything from distance seller Unigro. Both remotely and on credit. But how can you accurately assess the creditworthiness of a customer now that anonymous shopping has become so much easier with the internet? SAS and Python Predictions joined forces to develop a credit scoring tool for Unigro. This helps the company determine whether or not it should accept the order and its associated credit request. Over a period of two years, the tool has reduced credit risk by 20%.
The almost endless flexibility means you can just about do anything with it and expand your analyses enormously.
General Manager, Unigro
Specialized in microcredit
Unigro, a subsidiary of 3SI, generates 82% of its turnover from credit sales. They focus entirely on the niche market of credit consumers. “You can even buy electronic appliances from us on credit, in six instalments of 12 euros each, for instance”, says General Manager Yves Moens. “Our customers are people with limited financial resources, such as nest leavers or people who have recently divorced. A credit line for a small amount can offer them a welcome solution.”
While this microcredit is an asset for Unigro, it also represents a risk. “Our challenge is to grow our turnover without taking on more risk. A customer’s credit score is a gauge which helps us decide whether or not we can accept an order. Obviously, establishing an accurate credit score is vital for our company. We send our catalogues to prospects and customers who are deemed to be low risk, based on certain selection criteria, nevertheless anyone can shop online with us. Currently, 40% of our turnover is generated through online orders so credit scores have to be part and parcel of our company culture. When our marketing team devises promotions for example, they have to focus on managing risk as well as increasing turnover.”
Solid and reliable
Until 2004, Unigro’s credit activities were almost entirely managed by Cofidis, which is also part of Group 3 Suisses International. In addition, the company also applied a few simple models, which in the end proved insufficient to fully manage the credit activity. When Unigro decided to deal with credit aspects itself, the company had to find a professional tool that would enable it to optimally manage risk. “I had seen first-hand during training courses how solid and reliable SAS was for data mining projects. We quickly decided to develop a predictive model with SAS and Python Predictions, which specializes in predictive analytics.”
A credit score model was developed in less than a year. It attributes a credit score to each new order of a prospect or customer based on a selection of variables. “We started from scratch by determining all the factors that could influence credit risk. Initially we defined some 150 variables, which were then created in the system. Based on this, we moved to predictive models that use a limited selection of variables to predict credit risk. These include socio-demographic factors, such as age and profession, and we distinguish between new and existing customers, which takes into account past payment behavior. How many times has someone placed an order with us? Did the customer reimburse the credit line without problems? We have a lot of relevant data on this because we have very loyal customers.”
Scoring microcredit: a competitive advantage
This model has enabled Unigro to become a specialist in scoring microcredit. “Without this tool our acceptance levels (how many credit requests we accept and the extent of the underlying risk) and turnover would be far lower. However, we managed to maintain our acceptance level while reducing our risk by 20%.”
In addition, the tool also helped Unigro reduce the manual work involved in scoring potential credits to a minimum. It is easy when someone has a good or bad score: the credit request is simply accepted or not accepted. However, for some people credit risk is not a black or white situation. Sometimes orders stand out for some reason, for example because their amount is much higher than average. In such a case, the order has to be manually processed. An employee has to decide whether or not to accept the credit request. Their decision is then based on the customer’s file and a set of other parameters. Unigro has succeeded in limiting these manual interventions, yielding huge time savings.
From an update to a brand-new model
Since Unigro’s working and market context is also constantly changing, the model is updated every two years. This may consist of adding new variables that have become relevant or in changing the weighting for a variable. If a specific parameter is sufficiently crucial, this can even lead to new modules. The model to score fraud is a good example. “We now determine the fraud score for every new customer, followed by the credit score. The SAS solution provides support across the board in this respect. The almost endless flexibility means you can just about do anything with it and expand your analyses enormously.”
Also for marketing and forecasting
Yves is planning further regular updates in the future, especially for e-commerce. “The internet factor plays a stronger role in the scoring than was previously the case, as e-commerce is now a substantial part of our turnover. And the changes don’t stop here. The key is to identify which data are becoming more relevant, to test new ideas and to constantly re-evaluate the parameters. We know that we have the right partners to ensure that this will also go smoothly in the future!”
Unigro is extremely satisfied with its collaboration with SAS and Python Predictions. “Our employees in the credit department are very enthusiastic and firmly believe in the credit score model. We are building on the success of this model and currently developing a number of marketing and forecasting projects. We believe these projects will lead to improvements in all of our activities through the use of data mining!”
Develop a credit score model to identify the optimum balance between increasing turnover and limiting credit risk.
- Enables Unigro to accept the right credit requests and generate more turnover.
- Contributed to reducing credit risk by 20% over a period of two years.
- Reduces staff costs by reducing the percentage of manual interventions to a minimum.
- Do not be deterred by the initially unknown elements associated with a data mining project.
- Start by organizing a thorough consultation with your selected partner. We have a genuine partnership with SAS and Python Predictions.