Money Care credit broker and Odyssey Consulting Group have created a forecasting model based on Microsoft Azure Machine Learning. The solution assesses the probability of a positive bank response to a loan request. Money Care is the only independent credit broker in Russia not affiliated with any bank. The company was created in 2013 as a project of the consumer electronics and appliance retailer, Expert, to independently manage financing for customer purchases after it was revealed that the retailer was losing up to 25 percent of all customers who applied for consumer credit. Now more than 2,000 partners work with the Money Care platform, and the current pool of banks represents 90% of the market in the target segments.
Modern technology allows financial companies to work with a large set of data more quickly and efficiently, as it is a combination of different methods of knowledge discovery. For example, machine learning is a very comprehensive application of statistics to find patterns in data and create predictions of future behavior, outcomes and trends based on them. To increase the conversion rate of loan applications, Money Care decided to reduce the amount of questionnaire data to the minimum required, and to create a model that predicts the probability of a positive bank response. Money Care trusted determining the minimum set of data and the construction of the prototype to the experts in information-analytical systems with Odyssey Consulting Group.
"The reason for choosing the partner was simple - Odyssey Consulting Group is focused on understanding the customer's business and problems, being the driver of the latest IT solutions and technologies for the client. Using cloud solutions allows you to quickly deploy the desired infrastructure with minimal investment. Cloud technology opens up a wide field for experimentation and allows you to select the most effective variants of the most innovative solutions. For example, to use machine learning for forecasting without investing in the development of computing power or analytical tools". - Evgeny Lebedev, Business Development Manager, Odyssey Consulting Group cloud solutions.
The first phase for Money Care was a prototype classifier in Azure Machine Learning, whose task is to screen over 60% of loan applications with a probability of approval over 80%. The machine learning methods include: discriminant analysis, regression analysis, clustering, classification based on separability (SVM, ANN), and dimensionality reduction algorithms (PCA).
The second part of the project was to train Money Care employees on the principles of work and a joint workshop on improving the prototype. This phase included advice on setting up models in Azure Machine Learning, typical machine learning tasks, and identifying next steps to improve the prototype.
Despite the popularity of the topic, there are not many implemented machine learning projects. First, there is the poor quality of raw data - information that can be used for prediction is often simply not available. The second problem is the shortage of human resources.