Art of Analytics: Trapping Anomalies

Learn Data Science
Teradata Employee

TrappingAnaomalies-YasmeenAhmad-Web-650.png

 

About the Insights

This visualisation represents the detection of anomalous broker behaviours found by an insurance provider. The visual representation of the data highlights how quickly these anomalies become apparent when looking at connections in a graphical format.

 

The dots (nodes) represent quotes that are created by brokers using a platform provided by the insurer. Links between nodes indicate quotes that are associated, i.e. a broker used a previously generated quote (node) to build a new quote (linked node) after making some changes. Typical broker behavior indicates that once a broker has generated a quote, it would only be accessed and refreshed if the quote lifespan ends before a customer has taken a decision to accept the quote. The two clusters in the centre (bluish) depict anomalous behavior, where a broker is continuously returning and refreshing the same quote after changing a small number of attributes on that quote. This indicates the broker is gaming the insurer's system in an attempt to determine how the pricing engine works. This is undesired behavior and a fraudulent use of the insurer's system.

 

The goal of this analysis was to identify how broker's use the insurer's system and understand positive broker behaviours that lead to product sales. The aim was to identify how the system could be improved to support brokers and provide a better experience, as well as find preferential behaviours that support the insurer’s business and could be promoted to less successful brokers. This fraudulent finding was a byproduct of this analysis. The insurer can use this visual as evidence when holding follow-up conversations with the broker involved.

 

About the Analytics

This sigma visualization depicts analysis of data generated by a platform provided by an insurer for their brokers. This system logs all actions carried out by a broker on the platform. The initial part of the analysis involved identification of broker sessions on the platform and matching of sessions to a specific broker and customer. Within these sessions, this analysis focused on the quote related actions logged by the broker platform. These actions were captured and modeled as nodes.

 

Each node represents a quote generated for a customer in a distinct session. Links were created between nodes if the broker accessed the same quote and generated a refreshed quote in a new session. Graph analysis identified two large unexpected clusters of highly interconnected nodes that were anomalous from the other nodes in the dataset.

 

About the Analyst

Yasmeen is one of the most creative and insightful Data Scientists at Teradata. Yasmeen grew up in Scotland, where she enjoys the great outdoors, in particular hiking the Scottish Munros and sea kayaking.

 

Her work has seen her traverse many countries, including the UK, Ireland, Netherlands Turkey, Belgium and Denmark where she covers the finance, telecommunications, retail and utilities industries. Yasmeen specializes in working with businesses to identify their challenges and translate them into an analytical context. She has a unique ability to focus on how businesses can leverage new or untapped sources of data, alongside novel techniques, to enhance their competitive capabilities.

 

Yasmeen has worked with many analytical teams, providing leadership, training, guidance and hands-on support to deliver actionable insights and business outcomes. She uses various analytical approaches, including text analytics, predictive modelling, development of attribution strategies and time series analysis. She believes strongly in the power of visualizations and their ability to communicate complex findings to business users in a way that makes taking action easy.

 

Prior to Teradata, Yasmeen worked as a Data Scientist in the life sciences industry, building analytical pipelines for complex, multi-dimensional data types. Yasmeen also holds a PhD in Data Management, Mining and Visualization, which was carried out at the Wellcome Trust Centre for Gene Regulation & Expression. She has published several papers internationally and is a speaker at International conferences and events. In addition she has taught on MSc courses related to Data Science and Business Intelligence.

 

Yasmeen developed a keen passion for data analytics and visualization through her studies, having always been curious to ask questions and learn more. These skills have allowed Yasmeen to explore many opportunities in multiple disciplines, providing her with an endless world of new challenges!