AI Applied to Business - Net Sales Prediction

Two weeks ago, Arvato Systems successfully finished a three-day workshop together with Microsoft and STARBOARD Cruise Services. The workshop focused on how Artificial Intelligence (AI) can help in optimizing the business processes of STARBOARD. STARBOARD is part of the multinational conglomerate company LVMH (Louis Vuitton SE, Moët Hennessy), and is a premium retailer in the cruise line industry, focusing on revolutionizing the shopping experience on board of cruise ships. STARBOARD is operating on more than 90 cruise ships worldwide, representing ten cruise lines, such as Royal Caribbean and Norwegian, which is managed out of their headquarters in Florida. The challenge for STARBOARD is to not only provide the best possible customer retail experience on board but to also predict shopper behavior during the cruise. This includes net sales prediction, inventory management and other business operating topics.

STARBOARD's forecasting process is essential for their success as their transaction-based business model thrives on accurate prediction of net sales per passenger by voyage and category. Currently, STARBOARD does the forecasting manually on an annual basis which can take up to eight weeks. In these eight weeks, the prediction does not only take up time of resources, but it also requires a lot of subject matter expertise and previous observations which is dependent on STARBOARD’s current staff. Thus, the outcome of this process was based on individuals, their assumptions and not on standardized procedures, or fully leveraging accessible data. We at Arvato Systems have leveraged AI experts from our teams in New York and Guetersloh to change that. After meeting STARBOARD’s business and operation managers and understanding their pain points, we used unstructured data from STARBOARD’s previous years to build customized prediction models for their net sales needs by leveraging parts of the Microsoft Azure toolset. Starting with a small subset of the total 30,000,000 rows of 2015-2017 data, the algorithm and prediction model, we built, was able to show a pattern behind their data, and our forecasting report predicted a 2018 sales revenue within a $1 range of the actual numbers. We were able to build this first prototype within the two-day workshop which forecasts the net sales per passenger in only a few minutes. In comparison to the eight-week manual forecasting method, the automated method thus provides a dramatic increase in efficiency.

STARBOARD provided us an extract of a dataset containing their sales transactions of the individual shops they run on cruise ships. The goal in the workshop was to forecast the sales per passenger of a specific product category according to several parameters, e.g., the cruise line and voyage route of the ship. We thus aggregated the transactions according to the separate products as well as cruise lines, etc. In order to achieve our prediction goals, we trained an AI model. By upscaling our AI model to the complete dataset, we expect very high and robust accuracy. In accordance with our Microsoft partner, we host the model algorithm in Azure Databricks to provide full use of high-performance clusters. The data is transferred via API to Azure blob storage. Results of the model, as well as general data insights, are displayed in Microsoft’s Power BI. That guarantees an easy and straightforward visualization of the forecast and other information for the customer.

This workshop was a great door opener, and we are currently discussing with the company to see where else AI can improve business processes based on automated reports. For example, similar AI training and algorithms can be used for determining promotions based on the length and day of a cruise in addition to which category. E.g. High-priced items should be discounted in the first quarter of a voyage due to less spending of customers. Price and inventory optimization can also be an area of interest.