Increasing acceptance rate through machine learning for My Community Finance
30% increase in offer acceptance
Real-time model response in milliseconds
End-to-end implementation running fully on the cloud
The behavioural model improved the loan-offer acceptance rate of a digital channel using machine learning and real-time data processing.
We have chosen Profinit as our strategic partner for data analytics and machine learning as they show a very professional approach, excellent performance and profound knowledge. Thanks to our collaboration, we have learned valuable insights from the data and improved the acceptance rates of our offers by double digits.Petr Luksan
COO & Member of the Board
The UK fintech company My Community Finance, the most prominent lender in the UK credit union sector, was looking for a strategic partner in machine learning and data analytics.
The first step to achieving the ultimate optimised underwriting process with a personalised offer was to build a behavioural model to predict the probability of acceptance of each client quote with the given parameters.
Profinit accepted a request for proposal (RFP) by My Community Finance in the form of a contest to get the best prediction results from an anonymised dataset. Our data science team successfully tackled the challenge and delivered the best model out of all the competing vendors within two weeks.
The solution needed to meet the following specifications:
- Process hundreds of client features from the underwriting process and external risk to credit-bureau data
- Deliver high-precision predictions for thousands of quotes daily with minimal latency (milliseconds)
- Enable failover model retraining with a single click
- Increase the number of loan offers accepted
- Get valuable insights from quote data
The model needs to process hundreds of client features from the underwriting process and external risk to credit-bureau data.
Furthermore, the computational time is critical as each offer needs to be shown to the customer within a window of a few seconds when other competing offers are generated through web aggregator comparison services such as Experian.
Solution: Machine learning
Profinit designed and implemented the model for assessing each individual client quote. The behavioural model enhances the underwriting process by optimising offers for unsecured loan products using machine learning.
The end-to-end implementation consists of a real-time data processing pipeline running entirely on the AWS cloud and MLOps environment, enabling failover model retraining with a single click.
The solution provides stable, highly accurate predictions (85% AUC) and makes decisions in less than 100 milliseconds. The number of loan offers accepted increased by 30% as a result of using the solution for the individual offer for each customer.
The solution we designed and implemented has achieved these results for the company:
- We delivered a behavioural model to optimise underwriting with a personalised offer in real-time with a minimal latency.
- The model predicts the probability of acceptance of each client quote based on hundreds of features incl. credit-bureau data.
- The number of loan offers accepted increased by 30% as a result of the individual offer for each customer.
Would your company benefit from accessing a similar behavioural model?
Profinit improves the way organisations use and access data in-house. Let us show you how.
Related success stories and use cases
Equa Bank Propensity scoring model
Creating a prediction model aimed at identifying clients likely to take out consumer loan products and optimising campaign targeting.Learn More
Raiffeisenbank Competitor loans consolidation
Profinit helped Raiffeisenbank CZ detect twice as many loans with competitors – and approach more clients to consolidate their loans – while remaining “the most customer-friendly bank”.Learn More