Clients Churn ModelClients Churn Model - Payment processing solutions
The Client Churn Model is a predictive model designed and implemented in the financial sector.
Scope of work
With over 2 million global customers in 36 countries, our client’s company leverages the world’s best technologies for our partners – from large worldwide enterprises to locally owned small businesses, in more than 131 currencies.
The main challenge was the enormous scope and a large number of data to search, create and test in a short period, making quite a big team of IT specialists on-demand as long as accuracy tracking and updates for a significant amount of data.
Our main goals on this project were to deliver high-class service and to show the accuracy that we provide even working with a significant amount of data and a short time.
For data mining and preparation, we searched three different databases, and around 200 hundred variables were created & tested in a short time with the highest accuracy possible. Also, to test multiple predictive algorithms and choose the best one, we had to work with a dataset of a few million rows.
For managing and monitoring models in production, we had to track accuracy and keep up with the updates. In the final solution, we made for creating and introducing separate models for five markets. We had to deal with the automation of the whole process, from data import, through a model update, to data export into an external database.
Our client is a fintech company that has been a global leader in payment processing for over 30 years. With a broad range of technologies, their scalable payment solutions are tailored to meet the needs of any size business in any industry or processing environment. For that, they needed a solution that could make the most effective use of their existing technological systems and workforce to cope with the growing volume of digital customers while under pressure from the revenue issue due to low profitability and demand for loans. To do that, we had to deal with data mining and data preparation, as long as testing multiple predictive algorithms and choosing the best one.
As a solution, we’ve searched three different databases, and around 200 hundred variables were created & tested in a short time with the highest accuracy possible.
The biggest challenge we’ve met by working with such extensive data was little time. But because of our experience working for fintech solutions and working with data after dealing with some automation of the whole process from data import, through a model update, to data export into the external database, we could present our separate models for five markets.