Manufacturers are under growing pressure to put AI to work in their factories to catch defects in real time, guide operators, and extract additional capacity out of existing lines.  

However, many companies struggle to implement agents effectively on the floor, often struggling with the data that agents need to be effectively "trained." Companies often fall short in organizing their data; connecting data and agents across company teams; and handling the volume of data that is useful for training and implementing agents. 

See also: AI stokes debate over cloud-powered compute vs. on-prem  

Some are finding the benefits of data repositories in their organizations and are already using this strategy successfully.  

One company doing this is Durham, North Carolina-based Wolfspeed, a developer and manufacturer of wide-bandgap semiconductors that specializes in silicon carbide materials and devices for applications in transportation, power supplies, power inverters, and wireless systems.  

Repositories and foundations 

Wolfspeed officials credit their success in part to the company’s data repository—an "ecosystem" from where AI agents receive data—that holds information from manufacturing systems, operational documentation, troubleshooting logs, and engineering discussions within one hub.  

That ecosystem is a combination of structured data—such information from tool reports and factory floor settings—and unstructured data—such as institutional knowledge, information from company presentations and emails, among more. 

Wolfspeed uses architectural foundations for these agents provided by AI agent software service Snowflake Intelligence, which lent the architecture to unify Wolfspeed’s structured and unstructured data.  

Both types of data—according to leaders at Wolfspeed that Smart Industry interviewed—were integral in developing useful AI agents that employees across teams could use. However, understanding how to process and use company data to inform AI agents is a continuous process that requires trial and error.  

See also: Stories of AI adoption: Wolfspeed all-in with 22 agents across key company teams 

“A lot of data extraction for AI [involves] making sure that you have not only the right data, but you have the right governance across that data and the right architecture,” said Priya Almelkar, Wolfspeed's CIO. 

Robust data lakes, large repositories of raw information, draw from disparate sources of data spread across a plant or facility. The larger the breath of data gathered from across the manufacturing and supply chain processes, the better AI analyzes patterns and makes predictions. 

Drawing data from separate sources naturally lends itself to creating silos for data. To break down those silos, Brazil-based glass manufacturer Vivix employed the low-code AI development software Mendix to create linkages between these varied data repositories.

The resulting improvements demonstrate why cleaning and connecting your data is worth the time and investment. 

Building the foundation 

When Aristoteles Neto, industrial transformation manager at Vivix, took over the department in 2021, he focused on overall strategy, not technology. In 2022, Vivix began using low-code AI software to unify and clean the company’s manufacturing data—a complicated, frustrating task, at best. 

Podcast: With outdated and isolated OT, your AI strategy may be built on a blind spot 

“How can you build this data foundation, this foundation for using data, not only machine data, but production data, transaction data, documents and engineering data? The complexity of industry regarding data is very high,” Neto said. 

Neto pointed to two advantages when he began his campaign to improve how Vivix processed data. 

The mandate to create the industrial transformation department came straight from the C-suite. Neto therefore had wide latitude to experiment and looked to build a portfolio of solutions, not just a single pilot project. Vivix even implemented a new position, value management officer, to report the impacts of the projects. 

Neto also worked with plants that featured state-of-the-art automation. OT was not an impediment to creating the data foundation he wanted. Disconnected business layers created the problems.

“When I thought about how to build a digital solution … it’s difficult to use data from a machine and contextualize with different processes. If you use too many disconnected systems, you need a lot of different people to create your solution. Eight, sometimes more than 10 people from different departments like the AI expert, the generative AI expert, data engineers and automation engineers,” Neto said. 

Good data makes everything better 

Neto rattled off a series of benefits Vivix enjoys from analyzing normalized, contextualized and unified data: 

• 85% reduction in production issue resolution time. 
• Saved 6,000 work hours in a single year. 
• Increased energy efficiency of furnaces by 5%. 
• Avoided $1 million in maintenance costs in 2025.

“We increased our net promoter score (NPS) from our customers because [having a unified data architecture] allowed us to answer some complaints from our clients in two or three days at maximum. Before it was two or three weeks. Now we can understand our pain points from our clients and give better answers and make faster decisions, reduce the cost of operations and give some stability,” Neto said. 

Unified data means agility  

The unified data foundation created by Neto and his team supports experiments with agentic AI, digitalization, predictive maintenance and robotics. The more points of data, the easier to draw conclusions as to what innovations warrant further explanation or not. Even failure enriches the data pool. 

See also: Report: Manufacturers split on whether to prioritize industrial AI 

“We reuse this architecture, we reuse these foundations for projects in logistics, maintenance, production, security. … We use agile methodology. Sometimes we start a project thinking [there is value add], but when the project is [at the midway mark] and we understand that [the value add hasn’t] happened, we stop the project and put our effort into another initiative,” Neto said. 

“Now we can use any data that’s available in an easy way. It opens the door to creativity.”