Dynamic Pricing Under Uncertainty in Automotive Services Using Reinforcement Learning and Temporal Convolutional Networks

Dynamic pricing models for automotive care services require processes that reconcile short-run patterns of profitability and long-run strategic goals such as customer loyalty. Rule-based systems are inflexible to adapt to shifts in market conditions and do not optimize sustainable profits, especially under uncertainty in customer behavior and service demand. In this paper, we present a utility-based reinforcement learning (RL) model enhanced with temporal convolutional networks (TCNs) to learn time-varying patterns of maintenance histories and identify top pricing drivers. The TCN architecture was tuned best with a randomized grid search over 30 settings. For the sake of greater interpretability, we include a gradient-based feature attribution method that estimates the impact of inputs such as mileage, service interval and margin on price advice. Experiments on real-world automotive service center data indicate that our approach significantly improves predictive performance, reducing the mean squared error (MSE) to approximately 4.21%, with a coefficient of determination of 95.85% and a Pearson correlation coefficient greater than 98%. The combination of utility-aware policy learning, temporal feature extraction and model explainability results in a highly effective and interpretable dynamic pricing system for the automotive service business.