What are some best practices for managing forecasting accuracy for seasonal products like juices, water, etc.?

I would mention the following set of practices to manage forecasting accuracy for seasonal products:

1. Develop a strong master data management as a foundational practice.

2. Aim to do a proper data cleansing in order to improve as much as possible the quality of your data. Determine a frequency to do data cleansing in a regular basis.

3. The best time series to predict seasonal products in my experience is the Holt-Winters model. During the last year we have tried a new platform based on Machine Learning; we are still analyzing if this tool can help us to improve forecast accuracy for this kind of products vs previous techniques used.

4. Develop a tight connection between promotional and pricing activities with the Commercial team. Embed learning from other experiences in this regard into your forecast and monitor performance vs expectations.

5. Design and implement a strong collaborative process among functions involved on forecast generation; do not forget to consider rewards and recognition to incentivize the collaborative process. Measure the Forecast Value Added as a check that this collaborative process is working for your company.

6. Continuous improvement is a must. Try to understand what is the "failure mode" while losing FA and detonate corrective actions.  

One suggestion purely based on historical data - You may want to consider time-series methods like seasonal decomposition of time series (STL) or more advanced models like SARIMA (a combination of autoregressive AR models, moving average MA, and differencing). It will all depend on your historical patterns and how you incorporate them into the forecasting model to improve your accuracy.