Samsung Electronics Improves Demand Forecasting Using Amazon SageMaker Canvas

Based in South Korea, Samsung Electronics is a global company offering people around the world access to technology, such as mobile phones, computers, and smart devices. The Samsung Device Solutions division of the company focuses on the inner workings of electronic devices to provide maximum performance, reliability, and longevity.

Within the Samsung Device Solutions division, the Memory Marketing team analyzes memory needs for electronics produced by the multinational company. It previously forecasted memory chip demand based on customer preferences, external research, and simple regression. However, these inputs were sometimes volatile, inaccurate, and didn’t account for new factors. For example, with new applications and devices on the market and environmental factors like the COVID-19 pandemic impacting business, it became difficult to determine the inflection point by solely looking at previous trends. To overcome these challenges, Samsung Electronics sought a new methodology for demand forecasting. Rather than increasing the workload of its data science team, the company wanted to empower business analysts with no ML or coding experience to inform data-driven decision-making with ML using Amazon SageMaker Canvas, which provides business analysts with a visual interface for generating accurate ML predictions on their own, without writing code.

Samsung Electronics kicked off the project in April 2022. Then, in August 2022, it started training business analysts from the marketing intelligence group, a portion of the Memory Marketing division, through AWS Data Lab, which offers accelerated, joint engineering engagements between customers and AWS technical resources to create tangible deliverables. Five members of the team went through 5 days of training to learn how to use Amazon SageMaker Canvas. By September 2022, the business analysts were using Amazon SageMaker Canvas to analyze data and forecast demand over the next eight quarters for PC shipments.

To forecast demand, business analysts imported data from various sources, including internal data and external data from third-party sources, into Amazon SageMaker Canvas. After importing the data and selecting values to predict forecasting demand, Samsung Electronics could automatically prepare the data, explore it, and quickly build ML models. “All of these steps are done with a click, so business analysts can easily use the tool,” says Dooyong Lee, manager of marketing intelligence at Samsung Electronics. After building a demand forecast model using Amazon SageMaker Canvas, Samsung Electronics is seeing highly accurate predictions. “Using Amazon SageMaker Canvas, we can continuously advance the forecast accuracy over time,” says Lee.

By equipping business analysts with the skills to use Amazon SageMaker Canvas, Samsung saves time for both business analysts and data scientists. The marketing intelligence group meets weekly to analyze future demand for the company’s resources. In the past, it couldn't determine how a particular factor would impact demand on its own. “Using Amazon SageMaker Canvas, we can quickly see how a factor will affect the model,” says Lee. “Previously, we had to ask our data science team for help and would typically wait for 1–2 days. Now, we can save time by getting the answer using Amazon SageMaker Canvas in 1–2 hours.” The data science team can then focus on working with more advanced models, which is a better use of its expertise.

If the marketing intelligence group does need assistance with a model, it can collaborate with the data science team using AWS services. Business analysts using Amazon SageMaker Canvas can share the same model with data scientists who use Amazon SageMaker Studio, an integrated development environment that provides a single web-based visual interface for data scientists to access tools to perform all ML development steps. Using Amazon SageMaker Studio, data scientists can evaluate model results and parameters. “The data science team is small and has a lot of responsibilities analyzing advanced models,” says Lee. “It makes sense to have business analysts working with simpler models because we can still collaborate with the data science team if we encounter challenges.”