With its three well-known brands—Sub-Zero refrigerators, Wolf ovens and cooktops, and Cove dishwashers—Sub-Zero is widely recognized as a manufacturer of top-end home appliances. Founded in 1945 and still privately owned, Sub-Zero is not necessarily a company many would expect to be at the forefront of industry’s digital transformation.
At the Snowflake Data Cloud Summit, Justin Swenson, product technical lead at Sub-Zero, explained that in the six years he has worked at Sub-Zero, the company has moved from having 100% of its compute handled on-premises to having 70% in the cloud. The key to this massive move of computing power at such an established company “has been having a modern data stack,” he said.
Sub-Zero needs this “modern data stack” to manage data not just from its own manufacturing operations, but also the company’s connected products and its supply chain data connections. These data points provide critical information to Sub-Zero, such as component performance and most frequently used features that help the company predict maintenance needs and determine future product innovations.
Lacking a single, governed platform to unify data across its systems (prior to its use of Snowflake technologies), Sub-Zero couldn’t develop a complete understanding of its customer base or manufacturing operations. “Our previous systems weren’t going to sustain where we wanted to go,” said Jody McDonough, vice president of IT and CIO at Sub-Zero. “This [digital transformation] vision required us to clean up our data and make it more insightful—something we can act on.”
After testing different options, Sub-Zero chose Snowflake’s Manufacturing Data Cloud for its scalability, ease of use and support for semi-structured data.
Manufacturing re-assessment
Driving this move to the cloud for Sub-Zero has been the redesign of its manufacturing processes to “have better standards so that we can meet all of our time-oriented goals,” said Swenson. And, as has been the case for many manufacturers’ modernization efforts, COVID was a key accelerator for Sub-Zero as well. In Sub-Zero’s case, it was a matter of being able to keep up with a huge uptick in sales.
“2020 was a booming year for us in terms of orders that we couldn't keep up with because portions of our manufacturing operations were shut down at the same time as a lot of people went from eating at restaurants to staying home,” Swenson said. “And we've basically been trying to keep up and catch up ever since then. We recently caught up on a lot of our oldest orders to the point where we can now deliver product within a week, where at one point we were almost a year behind because we have so many orders.”
A critical goal for Sub-Zero in its manufacturing modernization is having the ability for its production operations to continue operating if internet connections drop or if any work cell or facility goes offline for any reason. The company also wanted near real-time monitoring of its manufacturing lines for ongoing performance analyses.
“We realized that a lot of our software applications weren't agile and flexible enough in terms of what needed to be collected or being able to adapt when supervisors or product line leaders made changes to any processes,” explained Swenson. “And we knew that if we had too stagnant of a manufacturing execution system (MES) that couldn’t adapt on the fly, our staff would stop entering data into that system.”
Proof of this was found during an operations assessment as part of the company’s update. Swenson noted that Sub-Zero found different places on the line where people had bought tablets and installed Excel on them to start tracking processes to avoid the company’s MES because it wasn't flexible enough. That’s why “being able to adapt while maintaining standards” became an important data management issue in the company’s ongoing digital transformation.
Transforming data management
The main shift for Sub-Zero in its move to the cloud involved moving from use of SQLServer on premises to Snowflake in the cloud using FiveTran to handle the ETL (extract, transform, load) processes.
Swenson noted that Sub-Zero uses multiple Snowflake offerings to track 100% of its production data today and uses Monte Carlo to monitor and analyze its data in Snowflake.
“You really have to look at what your ideals are as a company and what it is you're looking to do going forward from an IT strategy perspective,” said Swenson. “We had been customizing on premises, but as we moved to the cloud we started looking at more at multi-cloud SaaS (software-as-a-service) providers. And that opened us up to looking at many younger companies where before we looked at mostly established businesses.”
These multi-cloud SaaS options are important to Sub-Zero because they enable the company to keep its options open if it decides to move to other cloud providers.
“Our selection process for a lot of these tools came down to what we want our IT department to become, not just maintain where we're at today,” he added.
Data governance
The overarching goal of data governance, as explained by IBM, is to “maintain high-quality data that’s both secure and easily accessible for deeper business insights.”
At Sub-Zero, Swenson said they were focused on how data governance would contribute to the company’s manufacturing data mining process. At a basic level, this involves the ability to compare manufacturing metrics across facilities, manufacturing lines and teams, which requires a focus on how data is consistently collected and modeled for real-time analyses.
At the next level, you must have clear definitions of terminology, said Swenson. “When someone looks at one of our Power BI reports, do they know what every word on that report means? And if they don't, can they look it up? And it’s the same with metrics. Do I know exactly how this is calculated and where that data came from so that I can validate it was calculated correctly or be able to follow lineage to understand if the data is coming from the correct place and determine if it is a trusted source?”
A critical aspect of data governance in manufacturing at Sub-Zero centered on its MES master data, which is controlled by the company’s senior manufacturing engineers. “If a manufacturing engineer can go in and swap out operations, station IDs or change hierarchies, that could break or completely change downstream reporting,” said Swenson. “So we had to address our approval process over that master data. Yes, someone might have access to that MES and they might be the right person to change that MES system, but are they the right person to change how all the data and those relationships interact? And what kind of communication process or approval process should they go through to understand that this is an approved change that we all know about and understand and be able to validate the upstream and downstream effects of making a change?”
A step-by-step approach
Detailing the steps taken at Sub-Zero to implement its Snowflake cloud-based data management to resolve the questions noted above, Swenson pointed out these eight major activities:
- Define terminology across systems to ensure everyone is speaking the same language.
- Ensure metrics are calculated in the same way across facilities and identify the most important metrics. Though Sub-Zero, like any manufacturing company, often wants to do deeper analyses of its most important metrics, Swenson notes that IT and OT (operations technology) teams need to be careful when doing this because of the costs of cloud computing. “You're running compute every time you run a query [in the cloud],” he noted. “If you're running deeper data quality checks, you're running your costs up, so you need to understand: What am I doing this for?” To keep these costs in check at Sub-Zero, Swenson said they will typically look at their most important metrics in Snowflake using Monte Carlo to understand the company’s most important data tables, then determine if running compute time in Snowflake is justified.
- Define system owners and differentiate between domains and ownership. “For example, I have a manufacturing data owner, but I also have a manufacturing engineering data owner,” said Swenson. “So, who controls what and why, and what's the justification behind that?”
- Change approval process. Make sure that you have at least one such process, even if it's manual, advised Swenson.
- Automate processes. Sub-Zero focused on automating critical aspects such as security first. Swensen said, “We asked: How can we, in an automated fashion, make sure that someone doesn't have access to something at the click of a button and that any requests go to the correct person for approval?” With such security approvals now automated, these requests almost never go to IT now, said Swenson, but to a data steward or data owner who knows what this data represents and who in the company should be allowed to see it.
- Standardize data collection processes but keep these processes agile to address specific needs.
- Incident management observability. With Snowflake, Sub-Zero can now monitor its entire production pipeline. “Because we're putting everything into Snowflake, we can monitor it all with Monte Carlo to analyze our production pipeline,” said Swenson.
- Anomaly detection and incident handling. “It's not just that you can detect an error,” advised Swenson. “It's how you handle that error. Sometimes you can detect a problem and find out later it wasn't as much of a problem as you thought it was. Other times you detect something like a data table that hasn't loaded in five days.” You have to develop a process of investigating alerts when you see them to better determine what constitutes a real incident worthy of investigation.
Ongoing efforts
Since starting its latest data management assessment in December 2023, Sub-Zero now has a list of 24 major metrics it’s working through with its manufacturing teams. Swenson noted the company has already resolved how to handle 12 of these metrics and the process is expected to continue throughout 2024.
"As we work through these metrics, we have to look at what reports those metrics are used in. What feeds those reports? Where does that data come from?” said Swenson. “We're looking at the definitions of the terms used in those reports. We’re looking at the calculations for those metrics and the source information as well as our collection processes. Once we fully understand these factors, we will then have validated reports and validated data that can be trusted going forward.”
Swenson added that, as a long-established company, they’re having to deal with several long-standing processes to fully modernize their operations. For example, they’ve found that certain metrics come from an Excel file only stored on one person's computer.
“That’s a common thing we see all the time,” said Swenson, “but we all know it's a problem that’s got to go away. There's no use being upset about [the way things have been done at] a family-owned company that's 78 years old and where people are used to just getting the job done. But as you modernize, you have to resolve these problems as you find them.”
