Making SEPA Direct Debits more Secure

Challenge

Development and deployment of a fraud prevention system supported by AI to secure the SEPA direct debit procedure for online trading.

Solution

A microservice that is accessible via an API and determines the probability of default of a SEPA direct debit transaction, protects online merchants from fraudsters and thus minimizes losses.

Technologies

Python, C#, Microsoft SQL, Elastic Search, Tensorflow, Docker

 

Challenge

Consolidation of various internal and external data sources to analyze patterns of checkout behavior and identify potential fraudsters.

There is never 100% protection against fraudsters and insolvent customers. The costs incurred for bank processing, debt collection, manual process handling and the lost value of goods are immense.

It should therefore be attempted to reduce the quota of SEPA chargebacks to an acceptable minimum.

Client

The client is a full-service payment provider for all services relating to the billing of services and goods on the Internet. This includes the provision of payment systems and risk checks, processing and acquiring, sending invoices, monitoring incoming payments and chargebacks, debtor management, collection and reporting.

The aim is to further increase the efficiency of risk assessment using AI-supported techniques. In order to improve the quality of the risk checks, the client came up with the idea of using its financial data know-how and machine learning to further reduce the expenses for chargebacks.

Solution

A microservice that is accessible via an API and determines the probability of default of a SEPA direct debit transaction, protects online merchants from fraudsters and thus minimizes losses.

While potential fraudsters surf in a web shop, they leave traces like any other customer. The surfing behavior of these groups of people often differs from that of good customers, albeit only minimally. Previously, such patterns could only be identified with extreme difficulty and a great deal of manual research. Adjustments in the Fraud Prevention System were therefore always subject to a certain time delay.

In the preliminary discussion, the following individual steps crystallized out.

  • Analysis of the data sources
  • Use case definition and elaboration of the application areas of the system
  • Evaluation of new solutions
  • Formulation and testing of hypotheses
  • Selection of a suitable ML model and testing of its quality
  • Test of the prediction potential
  • Development of a risk service with API
  • Test and introduction into the production system

Throughout the entire process, great importance was attached to the protection of sensitive customer data. It had to be ensured that no data could be tapped at any point. Therefore, a semi-autonomous system was developed, which allows the client to coordinate which data is used for the calculation and how it gets into the new system.

The entire development process consists of 4 steps.

 

Stage

Data understanding and validation

Feature Engineering

Modelling

Deployment

Work

Collection, processing and validation of the customer’s data.

Convert raw project data to features.

Train the model on the basis of the prepared data set.

Delivery and provision of the model; provision of a user-friendly interface for the client to access the trained model.

 

Architecture between Client and Masasana

Result

A micro service that is integrated into the customer’s infrastructure and increases the quality of the fraud prevention system.

Developed by Masasana AI and integrated with MAI, the service provides access to the API that effectively assesses the probability of failure of a SEPA Direct Debit. Forecasts can be made in real time and are therefore a perfect tool for the use of risk assessments during a check-out process.

Cookie Consent with Real Cookie Banner