AI in Manufacturing Must Raise the Floor, Not Just Chase the Ceiling
February 19, 2026
GrayMatter Welcomes John Genovesi to Board of Directors in Partnership with Tailwind Capital
March 5, 2026AI in Manufacturing Must Raise the Floor, Not Just Chase the Ceiling
February 19, 2026GrayMatter Welcomes John Genovesi to Board of Directors in Partnership with Tailwind Capital
March 5, 2026AI in Manufacturing
Machine Learning in Manufacturing Delivers ROI: 5 Takeaways From Think!AI
How is machine learning improving things like quality control in manufacturing? Andrew Drake shares a few game-changing applications of machine learning in manufacturing.
For the last couple of years, manufacturing has been flooded with AI promises but not many applications that change how a plant actually runs.
What has changed operations — quality, waste and downtime — is machine learning in manufacturing.
At Think!AI 2026 in Pittsburgh, GrayMatter’s Andrew Drake showed where companies are finally seeing measurable ROI. It’s coming from machine learning models that analyze and interpret numerical data sets. These algorithms are helping solve everyday problems in manufacturing operations, from minimizing waste to predicting routine maintenance.
“We really need to see cold, hard, proven ROI in something that’s actually going to affect the process and what we’re making,” says Drake, Advanced Analytics Practice Director. “If it’s going to be a science experiment, usually manufacturing companies are not particularly interested.”
Check out five major takeaways from his talk on how machine learning is transforming today’s factories:
1 | Manufacturers have the data for machine learning. It’s just not AI-ready.
While data accessibility has improved, GrayMatter research shows that only 7% of our customers say they’re at a high level of data readiness. Drake believes the true number may be higher — yet major data challenges remain.
One struggle is historizing time series data: high-speed, sub-millisecond sensor data that establishes what “normal” looks like on the plant floor. Manufacturers often collect this data but don’t retain it in a usable historical record.
That gap is more common than many teams realize. “Every company has it, but a lot are not historizing it,” Drake says.
Once historized with the right context, the data can be used months or years later in machine learning algorithms to anticipate quality deviations and downtime. GrayMatter has been working to close this data gap for the last few years, Drake says, helping manufacturers structure plant floor data for machine learning deployment.
2 | Manufacturers don’t need AI experts anymore to use machine learning.
You don’t need to dive into code or understand a machine learning algorithm to deploy it on a plant floor, Drake says. Data formatting is more standardized, and that’s expanded who can leverage machine learning.
Early adopters in manufacturing relied on R&D to create and deploy machine learning algorithms. These teams typically had programmers and statisticians who spent a few years developing machine learning models before they delivered them.
“Now, process engineers on the plant floor and operators have the power in their hands to be able to predict something before it happens and develop that machine learning,” Drake says.
"Now, process engineers on the plant floor and operators have the power in their hands..."
3 | Machine learning can prevent bad product before it’s made, delivering real ROI.
Quality control in manufacturing can lead to unintended product and energy waste. Machine learning can reduce both.
Most manufacturing plants must meet strict specifications. Machine learning helps tighten and maintain those specs, reducing waste, increasing throughput, lowering energy use, shortening trace times, and improving scheduling, planning and predictive maintenance.
GrayMatter worked with a pet food producer with the goal of reducing waste. The kibble created was manually tested for fat and protein content and density. But if the product failed quality tests, everything produced since the previous passing test was discarded.
The team gathered sensor data from the extruders and oven, including oven speeds, zone temperatures and raw material inputs. “We have the data to predict what is your quality for those three things before it even comes out of the oven,” Drake says. “If it’s about to be bad, [we can] make suggestions to the set points of the oven to fix it.”
This machine learning algorithm helped the company save millions in waste and manufacture its products more efficiently.
4 | Without operator buy-in, machine learning stalls.
More people are using machine learning on the plant floor, but that doesn’t mean there’s widespread acceptance.
In the pet food use case, there was some cultural friction and pushback, Drake says. When GrayMatter started making real-time suggestions on what to tweak on the oven to fix quality issues, the operator who had been there for 40 years wasn’t entirely on board.
Machine learning doesn’t replace operators. Their experience matters, especially during outlier conditions and new product runs. But teams still have to connect the models to operator trust, typically by involving operators in tuning recommendations and validating results on the floor.
5 | Machine learning and AI aren’t top priority for many manufacturers — but that’s changing.
Manufacturers tend to take it slow when adopting new technologies, waiting for proven value before implementation. GrayMatter research shows the top plant-floor challenges include OT security, resources and people, IT/OT coverage and data strategy.
Predictive maintenance, machine learning and generative AI remain pretty low on their priority list, Drake says. But interest has grown significantly over the last couple of years.
“Even though machine learning was the hot thing five or 10 years ago, manufacturing is just now starting to standardize on it,” he says. “They’re starting to get true, hard ROI — millions of dollars in ROI — but it’s a journey.”
"To put a new technology
on the plant floor, we really need to see cold, hard proven ROI."
– ANDREW DRAKE
Advanced Analytics Practice Director