3M brings science to life through innovative products for home, business, and industrial customers. As a large, global company, 3M needed to standardize on an easily automated mechanism for sales forecasting to replace the multiple and highly manual methods in use across the enterprise. By adopting Microsoft Azure Machine Learning and other Azure AI services, 3M has been able to develop, train, and implement 1,500 custom models for each of its regions and divisions. These models deliver timely insights for finance, sales, and marketing employees and executives, while freeing them from time-consuming manual data manipulation tasks.
“We’re trying to push the boundaries and stay up to date with the best products and best practices in the industry. That’s what led us to Azure AI.”
Jeff Neilson, Data Science Manager, 3M
Innovation both inside and outside the company
3M is a company that never stands still. Whether it’s introducing iconic consumer and business products like Post-it Notes, delivering pioneering industrial components, or meeting the need for N95 respirators during the COVID-19 pandemic, 3M solves real-world problems and improves lives through applied science. This includes continually adopting the best possible technologies to optimize, standardize, and automate internal business processes so employees can focus on bringing the company’s innovations to customers around the globe.
For Data Science Manager Jeff Neilson and his six-person corporate finance data science team, this meant establishing new and improved ways to compile, calculate, and deliver sales forecasts across the company’s many regions and divisions. “It’s a challenging task because we’re a big, international company, and our sales teams were sometimes using different methodologies—often with a manual component—and generating mixed results,” explains Neilson.
Machine learning in the cloud for better forecasting
The team felt it was important to build the integrated forecasting solution in the cloud using machine learning capabilities, and Microsoft Azure was the platform of choice. “We’re trying to push the boundaries and stay up to date with the best products and best practices in the industry,” says Neilson. “That’s what led us to Azure AI. Our Finance department had already moved a lot of on-premises data to Azure, including a Snowflake data warehouse, and that facilitated our decision.”
Neilson’s team is using Azure Machine Learning, Azure automated machine learning, and the Many Models Solution Accelerator, which offers the capability to train and score up to hundreds of thousands of machine learning models in parallel. For 3M, the scope of the challenge was to work with about 1,500 models based on sets of time-series data for different region-division combinations, generating a custom model that’s fit for each data set.
“The Many Models Solution Accelerator was an extremely useful starting framework for us to build around, and it helped us deliver our first iteration of the solution in just three months,” says Neilson. “Because we worked in Azure, we could provision compute clusters and nodes as necessary to scale up and complete training jobs faster, which shortened our overall development cycle time.”
Collaborative planning and promising results
In consultation with its business partners, the team began with a thorough analysis of the business processes involved. The aim was to ensure the solution wouldn’t be just technically sophisticated but would solve the problem at hand. “It was important to have business buy-in so that we could develop the best possible solution,” says Neilson. “We also had great support from Microsoft throughout the process.”
The team built an Azure Machine Learning pipeline that learns patterns in past and current sales data to generate future sales predictions. In addition, they collaborated with another 3M group focused on integrating Microsoft Power BI across the organization. This helped them better compile and visualize results.
“For our executives, the improved forecasts and reporting provide an unbiased and unfiltered view of data that they may have gotten anecdotal reports about before,” says Neilson. “It offers them a great starting point for discussions and decision-making, as they figure out what is working well and what may need to be adjusted.”
The feedback so far has been positive. Neilson expects that 3M will have access to more frequent and accurate sales planning targets in the future than the company has today. Employees are also spending time more effectively. “As more of our groups rely on the Azure Machine Learning solution, our finance experts can focus more on higher-level tasks and spend less time on manual data collection and input,” says Neilson. “With Azure AI capabilities, business conversations become more meaningful, and strategic decisions become more targeted.”
“As more of our groups rely on the Azure Machine Learning solution, our finance experts can focus more on higher-level tasks and spend less time on manual data collection and input.”
Jeff Neilson, Data Science Manager, 3M
Follow Microsoft