Sponsored
Chemical Plants Don't Need a Data Cleanup to Unlock AI
April 29, 2026
3 min read
shutterstock_1637823094
“The asset data you have is better than you think,” said Marty Dytrych, business development lead for Bolo AI, a tech startup based in Palo Alto, California.
It’s a common refrain among manufacturers deploying artificial intelligence who believe their existing data isn’t AI ready, Dytrych discussed during his session on asset performance management with AI at the eChem Expo in Kingsport, Tennessee, April 8.
That's because they've been told by platform vendors they must create a “knowledge graph,” a process that involves mapping, tagging and labeling data before they can deploy AI. It’s a process that can take years to complete and cost millions.
But the data sitting in a company's CMMS, P&IDs, work orders, historians and other systems may be closer to being AI-ready than many chemical processing operators believe. A different concept, called a “context graph,” provides insights without requiring a time-consuming data cleaning process.
Dytrych described a different type of AI agent that can work with unstructured data and is already familiar with key industrial identifiers, such as historian tags. These "reasoning agents" use the context graph to understand what that data means and work through problems to produce outputs like work orders or maintenance alerts.
For example, Bolo AI is working with a chemical manufacturer that needed to replace retiring contract maintenance workers. The company was spending $2 million a year for 10 maintenance contractors, Dytrych said. The chemical plant didn’t want to invest in 10 more workers while losing the knowledge from its departing contract maintenance planners.
The company wanted to use AI agents to cut its on-site maintenance staff to three people. The planners were working off several spreadsheets, each with eight to 10 tabs. Bolo AI used its context layer to sift through roughly 2,000 internal photos, a spreadsheet with 200,000 to 300,000 lines, and a history of past work packages printed as PDFs, none of which had to be cleaned or mapped. Bolo AI describes its context, or semantic, layer as a type of business glossary and rulebook that sits between AI and existing systems.
“We were able to provide context around all their applications that they use from SAP to P6, their TotalView, which is the digital twin software that they use, all the way to building out a weather API so they could know, ‘Hey, we're planning something on this day. Be careful of this weather,’” Dytrych explained.
The process took about four to five weeks versus a year that it would typically take to map the data, Dytrych said.
The feed provides the chemical plant with outcomes that maintenance planners can use to make more informed decisions. Traditional large language models would require a person to prompt or ask a question, Dytrych explained.
“Today, what we're seeing is that you can deploy agents to work on your behalf, produce outcomes and then you can manage those outcomes and those agents as they do activity,” he said.
In the example Dytrych provided, the agent assembled five work packages from prior work orders, eliminating a significant number of duplicate work orders. The agent would determine if a work package was a duplicate, conduct additional research on contractor hours and materials from historical records, and provide its reasoning beneath each section. The three remaining planners would then review and approve before submitting to schedulers.
“It allows these reasoning agents to operate through the context to understand the how, so that when it grabs that data, it knows what it is, and then you can work through a reasoning journey to produce the outcomes that you want,” Dytrych said.
About the Author
Jonathan Katz
Executive Editor
Jonathan Katz, executive editor, brings nearly two decades of experience as a B2B journalist to Chemical Processing magazine. He has expertise on a wide range of industrial topics. Jon previously served as the managing editor for IndustryWeek magazine and, most recently, as a freelance writer specializing in content marketing for the manufacturing sector.
His knowledge areas include industrial safety, environmental compliance/sustainability, lean manufacturing/continuous improvement, Industry 4.0/automation and many other topics of interest to the Chemical Processing audience.
When he’s not working, Jon enjoys fishing, hiking and music, including a small but growing vinyl collection.
Jon resides in the Cleveland, Ohio, area.
Latest from eChemExpo
April 28, 2026
March 23, 2026