Fraud Detection at Fuel Sales Automation System

about
Our client is one of Turkey's fastest growing oil companies over the past ten years and currently is the industry leader in market share
Problem
  • Among the sales realized through OTOBIL (Fuel Sales Automation System), there was no adaptive and smart detection method for fraud or suspicious fuel purchases.
  • There was no infrastructure to systematically detect fraud cases based on data analysis.
  • Inability to quantify the number and amount of the total fraud and technical problem related cases that have occurred.
Action
  • Defined and categorized past suspicious transactions as Fraud or Technical Problem types, such as User Fraud, Internal Fraud, Third Party Fraud or Device Theft, Device Cloning, Duplicate Selling, etc.
  • Developed analytical models and expert rules to detect each type of suspicious transaction.
  • Created alarm lists with automated near-real time running models.
  • Developed Quality screens in order to improve the workflow process (i.e. detecting, measuring and monitoring the quality of alarms and fraud detection performance).
Tool Stack
  • Oracle, Phyton, Power BI

95

%

Improvement of preventing illegal fuel purchases

50

%

Reduction in case resolution time

30

%

Increased trust in the OTOBIL system and customer satisfaction