Fraud Risk Management Platform

GBG provides sophisticated Machine Learning models to optimise rules management, assess risk and enhance fraud detection.

Machine learning has many use cases, but broadly speaking they are used to either make predictions or for model optimization (Automated re-training).
Model Predictions complement the existing fraud rules, adding a risk-rating mechanism as well as detecting new fraud.
Fraud Rules are still a valuable tool as they provide a clear avenue of investigation for fraud reviewers.

GBG can offer a number of different model types and customers can use a combination of these to suit a wide variety of use cases.
Supervised Machine learning can be used to provide a mechanism to learn from confirmed cases. Unsupervised learning can also be used to provide anomaly detection and identify and underlying patterns and inconsistencies, without the need for confirmed cases.
We also provide the ability to design and manage custom models, giving a framework for customers that already have dedicated machine learning teams and capabilities.
Custom model creation can import open source libraries for future proofing and provide a notebook for further model analysis and direct Python coding.

GBG deliver machine learning via an open, user-controlled interface which enables both business users and expert users to, train and throttle the score threshold. This provides our customers with optimal detection and alert volumes based upon the organisation risk appetite and resources. Our machine learning module is integrated with the compliance investigator case management. Patterns of behaviour are automatically fed back to the machine to automate model training, thus ensuring continual adaptation and protection to evolving patterns. The module provides access to a growing list of machine learning algorithms which have proven effective in even the most challenging detection assignments. Our module provides testing, audit and tracking of changes, with contributing features being displayed to the investigator along with the score to provide model transparency compliance. Automated training, adaption and updating of the models can take place over specified rolling time window. Models can also automatically be published when they meet a given performance threshold.

The nature of application fraud and risk is both complex and continuously evolving.
As a consequence, fraud detection and risk assessment software needs to be equally as adaptable.

The best solution is one that can leverage a rich variety of layered data sources and combine that with layers of intelligent scoring. This powerful combination provides the tools required to deliver confident, accurate risk assessment and combat both common fraud typologies and new attack vectors.

The ability to easily design bespoke onboarding journeys via a ‘drag and drop’ user interface, complement application data with supplementary data sources and build a variety of scenarios using business level adaptive rules, score cards and machine learning provides a flexible tapestry for efficient onboarding.

The simple message for successful onboarding via Instinct Hub is:
Each layer adds more context to the decision. Each layer adds further accuracy and power to the detection. The more layers, the more value.