In used cars auction systems, users can buy vehicles through fixed-price rounds or participate in auction rounds where they place bids, with each item typically awarded to the highest bidder. This auction setup presents a challenge for recommender systems, as it involves sequential recommendation of unique items, where each item is available for sale only once in both fixed-price and auction rounds. Although this scenario is highly relevant, it has received limited attention in existing sequential recommendation research. Moreover, this challenge relates to the cold start problem encountered by many recommendation models. In this work, we aim to address the unique item sequential recommendation problem by developing an attribute-aware model for next-item prediction. Specifically, we introduce the Attribute-Aware Sequential Recommendation Model (ASRM), which is designed to handle unique item data and effectively leverage item attributes in the absence of item IDs. To further enhance performance in this context, we propose an improved version, ASRM++. Our experiments, conducted on a dataset from Volkswagen Financial Services’ used car center, demonstrate that ASRM significantly outperforms existing state-of-the-art models for unique item recommendation. Additionally, we present A/B test results from the deployed ASRM model to validate its effectiveness.