AI Engine Boost with GENAI-Synthetic Data Qorus-NTT DATA Innovation in Insurance Awards 2025

This project leverages synthetic data generation as a strategic tool to enhance operational efficiency and improve the accuracy of existing data-driven models. An innovative approach has been adopted to elevate the performance of traditional AI models and minimize prediction errors caused by data imbalances. In the first phase of the project, synthetic data integration was implemented in the Next Product Offer model to improve customer segmentation and recommendation engines. By generating high-quality synthetic data that mimics real customer behavior, the accuracy of the model was enhanced, resulting in more precise and profitable customer offers. In the second phase, the focus shifted to fraud detection models for both motor and non-motor fraud. One of the biggest challenges in these areas is the low occurrence of fraud cases in real datasets, which causes models to underperform due to imbalanced data distributions. To address this, synthetic data specifically designed for fraud detection models was generated using Generative Adversarial Networks (GANs), optimizing the system’s false positive and negative rates. Synthetic data generation has gone beyond traditional data collection processes to create a broader, more diverse, and secure training environment. Additionally, it has reduced manual labeling costs, sped up operational processes, and made the models more adaptable. This project positions itself as a scalable and sustainable data innovation solution that strengthens traditional AI engines, optimizes customer engagement, and offers significant improvements in risk management. With synthetic data production in the NPO model, the accuracy of recommendations made to customers in the pilot phase increased by 8%, resulting in a gain of 20 million TL and will lead to a leadership position in smart sales solutions.