How Artificial Intelligence can transform Insurance in Tunisia: Optimizing Underwriting, Enhancing Claims Management, and Strengthening Fraud Detection

Tunisia's insurance sector faces operational inefficiencies, high fraud rates in motor third-party liability (MTPL) with loss ratios exceeding 100%, and outdated mortality tables like TD 99 creating profitability and mispricing risks. An explanatory sequential mixed-methods design was employed: a survey of 56 insurance professionals using TAM, TOE, and DOI frameworks, followed by quantitative modeling with GLM for non-life pricing, Kaplan–Meier for mortality/lapses logistic regression for lapse prediction, Random Forest for fraud detection, and a claims-handling chatbot prototype. GLM outperforms manual tariffing for MTPL premiums. Observed mortality is 17–18% below TD 99 (Chapter 6). Lapse prediction achieves AUC ≈ 0.96 (95% CI: 0.942–0.978) [Table 16]. Fraud detection yields AUC ≈ 0.805 (95% CI: 0.801–0.898) [Table 24]. Chatbot reduces claims cycle times by 40% [Table 18], generating estimated gains of TND 2,029,750 (~USD 648,880) [Table 43]. A gradual "assist before automate" AI strategy enhances pricing accuracy, anti-fraud effectiveness, and claims efficiency. Priorities include data governance, model explainability, tariff updates, experience-based mortality tables, and secure data-sharing platforms. Future research should validate survival-based lapse models, behavioral variables, and production pilots.