Wealthon and Risk Prediction of Loan Defaults with Machine Learning
Deployment date: 2025
Sector: finance
Google Cloud
As part of optimizing customer service processes, Wealthon was looking for a complementary solution that would enable additional risk assessment at the loan application stage, including predicting repayment difficulties for customers using the POSCASH product. OChK proposed building an analytical feature repository and predictive Machine Learning models based on it. This allowed Wealthon to transform its raw data into actionable business insights. Thanks to collaboration with OChK, Wealthon has shortened the time needed to identify at-risk loan applications and continuously adapts its credit policy to the risk factors indicated by the model and to customer needs, ensuring mutual financial security.
About Wealthon
Wealthon is a Polish fintech founded in 2019 with the mission of supporting small and medium-sized enterprises through non-bank financing. Today, the company is building an entire ecosystem of services that provide SMEs with comprehensive support – from loans to business management tool. Its flagship product, POSCASH, is a loan where creditworthiness is assessed based on payment terminal turnover, with flexible daily repayments. Wealthon was the first to introduce this type of product to the Polish market.
Challenges
In connection with the optimization of customer service processes and the growing popularity of the POSCASH product among SMEs, Wealthon wanted to build an additional risk assessment indicator based on machine learning. This would enable the company to:
provide real support to the risk department by equipping it with an additional tool for assessing and validating POSCASH loan repayments,
continue developing its services while expanding its knowledge of the causes of default and how to identify them at the application stage, thus tightening up the entire process.
The Wealthon team wanted to consciously and effectively use its collected data to improve business operations. To deliver the project, Wealthon needed a partner to guide the process. OChK’s expertise in building data/AI solutions and its strong cybersecurity and compliance knowledge made it a natural choice. OChK proposed building a predictive model using Machine Learning methods.
Solution and Execution
To deliver the project, OChK experts and the Wealthon team defined the concept of default and identified the target group for analysis – POSCASH clients where defaults could already occur technically. The project was executed as an MVP and included several phases:
Analytical workshops – to confirm business goals and select the scope of data to be processed by ML models,
Data import and aggregation – using Dataform, OChK built a transformation matrix for the selected target group, preparing data for modeling (including aggregations such as sums, averages, and trends across different time periods),
Building and testing ML models – leveraging Google Cloud’s Vertex AI and BigQueryML platforms, the team applied multiple analytical techniques to generate predictions for both historical and new credit applications. The models were validated on external test data to confirm stability,
Scoring and automation – OChK implemented micro-batch scoring on live data and delivered business insights to Wealthon’s teams responsible for monitoring and preventing defaults.
The entire project – from problem definition to delivering insights and implementing scoring mechanisms – took only 10 weeks.
Used Technologies
BigQuery
BigQuery ML
Dataform
Vertex AI
Workflows
Results
With predictive models built by OChK, Wealthon can now identify customers with the highest default probability (top 20%) at more than twice the accuracy compared to previous approaches. Short-term impact: the ability to quickly flag loan applications at risk of default. Long-term impact: valuable business insights for shaping credit risk policies.
Wealthon now has an analytical base of more than 900 features, which can be used both for the current ML models and future predictive projects.
Automating data preparation and credit application scoring has significantly reduced the time needed to identify high-risk applications.
Risk assessment in the financial sector plays a key role, both in terms of portfolio security and stability, as well as the further development of the services offered. That is why the introduction of an additional solution, based on an ML model and the collected data, was particularly important for us. Cooperation with OChK in this area is a great example of how quickly projects can be implemented when two well-prepared and professional teams meet. I definitely recommend it!
Artur Milewski
CTO, Wealthon
What are your challenges?
Let's face them together!