Book your personalized demo
See what your data can do
can do for your business
can do for your business
Learn how to get started and what results you can expect.
Contact us
How predictive analytics helped an electronics retailer detect bonus abuse
A major electronics retailer lost money through the misuse of its loyalty program. Cashiers charged purchases to their own cards to collect bonuses. Manually reviewing videos and receipts was slow and inefficient.
Service:
Predictive analytics
Timeline:
June 2025
Country:
Ukraine
Industry:
retail
Company:
NDA, Electronics Retail
Summary:
Internal abuse is difficult to detect because individual transactions often appear legitimate. The merchant switched from manual selection to data-driven prioritization. We combined transaction and loyalty card data and trained a model. The security team now focuses only on the highest-risk cases.
Business area
The company operates a network of electronics stores with a loyalty program. Any misuse leads to financial losses and distorts the analyses.
problem
Fraud detection was inefficient. The team reviewed over 200 videos monthly based on intuition. Only 5 to 10 of these cases confirmed the fraud. The process was time-consuming and missed many incidents.
Solution
Step 1: Merge historical data
We combined data from separate systems. This included transactions and products, as well as payment methods and returns. Loyalty card activity patterns and locations were also integrated.
A unified view that combines historical records with AI predictions for every customer (anonymized)
+
Step 2: Create a training dataset with labels
We used previous decisions by the security team as learning examples. The system learned which patterns led to confirmed fraud.
Step 3: Train a fraud detection model
The model estimates the probability of bonus fraud. The goal is to identify cashiers using unauthorized cards. The result is a ranking for the security team.
+
Step 4: Optimize for limited historical data
Since there were few confirmed cases, we focused on creating the best list for review. Each review provides new examples and makes the system more accurate.
Results
- Only 50 instead Previously over 200 checks per month.
- 70 % Of the cases examined, 100 are confirmed cases of fraud.
- 35 cases of fraud were found in 50 videos.
- 25 hours Time saved per month for the security team.
The security team is now reviewing fewer transactions, but simultaneously detecting more fraud. This saves time and speeds up investigations. Each review generates new flagged examples, causing the fraud dataset to grow faster. Over time, this improves the accuracy and coverage of the scoring model.
