Salesforce relies on precise and correct sales forecasting to optimize earnings predictions, inventory planning, and strategic decision-making. Traditional deep learning models, such as LSTMs, RNNs, and Transformers, have been broadly used for time series forecasting in Salesforce but frequently struggle with capturing complicated and intricate sales patterns. To address this, we introduce SalesCoST, a original time series representation learning framework designed to enhance forecasting accuracy within the Salesforce ecosystem. SalesCoST employs contrastive learning techniques to disentangle seasonal and trend components in sales data, leveraging both time-domain and frequency-domain losses to learn sturdy and resilient representations. Our approach enables more dependable and trustworthy sales predictions, improves pipeline visibility, and enhances decision-making in CRM workflows. Experimental outcomes on real-world sales datasets demonstrate that SalesCoST surpasses existing deep learning approaches by 21.3% in MSE, making it a extremely effective solution for Salesforce-driven forecasting.