Case Study: Ferratum Bank

Ferratum Bank logo
Real-time lending decisions with ScienceOps


Banking, Finance


1,500,000+ Customers


20+ Countries


Real-time credit scoring and fraud detection


Ferratum (FRA:FRU) is Europe's leading mobile bank and provider of consumer credit and microloans, revolving lines of credit, and installment loans. As a thought leader in financial technology, Ferratum has pioneered European microlending since the company's inception in 2005.

Headquartered in Helsinki, Finland, the company has served over 1.5 million customers worldwide, with lending operations in 20 countries, including the United Kingdom, Finland, Spain, the Netherlands, Denmark, Australia and New Zealand among others.

"The time it takes us to deploy our models has gone from several weeks to just one or two days"

- Scott Donnelly, Director of Business Lending at Ferratum



Increase in loan approval rates


Decrease in loss rates


Increase on overall margins


Ferratum uses advanced statistical and machine learning models to make better lending decisions, detect fraud more accurately, and expand their customer base more effectively than other lenders. As the company began to expand into new markets, its need for new machine learning algorithms and specialized predictive models grew as well.

Scott Donnelly, Director of Business Lending, realized the data science team would have to deploy more complex models across a growing number of business functions. Searching for a platform for streamlining the process of moving from data science insight to production apps led Scott to Yhat.

"Our modeling tool of choice is R," explains Scott. "But we found a lack of dependable tools to deploy models to production with the reliability and support we require." Scott says his team evaluated various open-source R server tools but couldn't get comfortable with any of them. "We wanted a simple-to-use commercial option with good support to deploy R models natively, and ScienceOps fit the bill."


Eliminate IT overhead

ScienceOps streamlines Ferratum's model deployment process and eliminates considerable engineering overhead previously needed to monitor and maintain their production data science routines.

"The time it takes us to deploy our models has gone from several weeks to just one or two days," says Scott. "And by putting the bulk of the deployment responsibility into the hands of the analytics team, we've reduced the need for outside IT resources substantially. This means faster turnarounds, less bureaucracy and less room for errors in cross-environment model validation."

Ferratum Bank workflow using ScienceOps


Previously, Scott and his team ported their models from R to production systems manually, hard coding each model by hand. "Now," he says, "Anything we can model in R we can easily deploy into production natively using R. So, the breadth of what we can reliably put into production in a timely fashion has expanded greatly."

Transition to ScienceOps

Ferratum is already using ScienceOps to power underwriting and decision-making systems in 4 of its 20 international markets so far. This includes both customer-facing web and mobile apps such as ferratum.se as well as internal systems used by the company's front-line underwriting employees.

"Currently," Scott explains, "we're migrating all our credit scoring models over to ScienceOps. Plus, we're now using ScienceOps for offline scoring for things like customer lifetime value, customer segmentation, and marketing optimization. So we're using it for more than just real-time predictions."

Service and Support

According to Scott, Yhat's service and support is "excellent." He notes, "We get very fast responses to questions, and Yhat has actually developed several new features based on our suggestions which has been extremely useful to us. Overall, we are very happy with the choice to use Yhat ScienceOps!"

mobile loan application

Real-time risk and credit models embedded in a Ferratum mobile site


In a recent model deployment in Eastern Europe, Ferratum’s data science team estimated that by utilizing ScienceOps, they achieved:


Increase in loan approval rates


Decrease in loss rates


Increase on overall margins

By getting changes into production faster, Ferratum was able to realize both approval rate improvements and loss rate reduction 4 weeks sooner. These changes have outperformed the benefits of previous model changes in the same country utilizing the previous scoring platform.

Interested in learning more?