Artificial Intelligence for better sales forecasts at SMA Solar Technology AG

Challenge

As a leading global specialist for photovoltaic system technology, SMA offers solutions for private solar systems, commercial photovoltaic installations up to solar power plants in megawatt range. With a large portfolio of system technology and energy solutions as well as global target markets, a reliable sales forecast is a key element for SMA’s sales, resource and material planning. The previous forecast process required a great deal of manual effort and resources. The information basis for the responsible employees was primarily a retrospective view of selected internal data such as historical sales. Therefore, SMA identifies great potential in increasing the accuracy and process efficiency of the sales forecast.

Goal

SMA wants to use machine learning to increase the quality of sales forecasts and support sales.

Solution

SMA has commissioned data science experts from eoda to implement the AI approach in their sales forecasts. In order to achieve an optimal forecast for each product in each target market, eoda has chosen an approach in which the machine learning models, they have developed, adapt automatically to the individual product/market combinations and their specifics.

eoda also considered other internal and external data sources in addition to historical order histories in the forecast.

Result

Early in the proof-of-concept phase, the machine learning models were able to achieve a significant improvement in the accuracy of the sales forecasts. Specifically, the machine learning model reference values are up to 35% better than existing expert estimates (metric: MAPE – Medium Absolute Percentage Error).

These results support sales and serve as a better planning basis for a wide range of business areas. This scalable eoda approach with customized machine learning models can be rolled out to the entire portfolio and all target markets in a short time after the successful PoC.