How to Scale AI in Manufacturing

Most manufacturers know they need to start with AI. The common advice, “start small” with a use case, prove value, and scale, is incomplete on its own. Starting small without first thinking big leads to a collection of disconnected experiments that never compound into enterprise capability.

At IIoT World’s AI Manufacturing Day 2026, a panel featuring Conrad Chuang, Industry Marketing Consultant at InfluxData, Jonathan Alexander, Global Manufacturing AI and Advanced Analytics Manager at Albemarle Corporation, and Felix Strenger, Senior Expert Consultant Lean Production/i4.0 at Bosch, described a framework that sequences industrial AI investment: define where you want to be in 10 years, then execute the first concrete use case in 30 days.

Why a 10-Year Vision Must Come Before the First Use Case

If you want to build the Empire State Building, you do not hand people hammers, nails, concrete, and rebar and say “come back in 90 days.” Albemarle spent a year before scaling industrial AI globally, mapping every use case the organization wanted to solve over the following decade: process optimization, yield improvement, equipment reliability, OEE, energy efficiency, and sustainability. That exercise determined the technology infrastructure, the ontology design, the organizational structure, and the roles and responsibilities needed to support all those use cases. Once the direction was set, each subsequent step moved the organization toward it.

The organizations pulling ahead are the ones that treat AI as an operational capability rather than a project. They think about where they want the program to be in 5 or 10 years and work backward to determine what the technology, organizational structures, and processes need to look like.

Execute in 30 Days, Build Momentum

Once the direction is set, execution should start immediately with one high-impact pain point. The first 30 days should produce something concrete: a solved problem, a working prototype, a measurable improvement that the organization can point to and say, “Here is what is possible, and we did it in one month.”

That concrete result builds internal cohesiveness and gets people invested in the program. It also reveals something documentation cannot: the real constraints only surface when teams work with actual data, actual systems, and actual telemetry.

From there, the approach is incremental. Each subsequent use case builds on the semantic models and data infrastructure established by the previous one, creating a reusable foundation rather than another isolated experiment.

More Data Does Not Automatically Create More Value

A common assumption is that more data collection leads to better decisions, but the relationship between volume and value is not proportional. Organizations can accumulate vast amounts of operational data and create no additional value from it.

Data collected without a clear purpose becomes noise; data collected with a specific problem in mind becomes an asset. However, the opposite extreme is equally dangerous: collecting data at too low a resolution or too slow a sampling rate can mask the very issues the organization needs to detect. Transitory events and fast system dynamics disappear entirely if the measurement frequency is too low.

High-resolution data should be collected intentionally. Not everyone needs to consume it all, but it should be available for the people and systems that require it for problem-solving and analysis.

The Next Frontier: Collective Intelligence Across the Value Stream

Once an organization has mastered response speed and built trust in AI-assisted decisions, the next frontier extends beyond individual plant optimization. The vision described in the session moves from single-plant intelligence to collective intelligence that spans the entire value stream and supply chain.

A factory that dynamically adjusts its own processes in real time to meet changing demand. A supply chain where a delay in one region automatically triggers production schedule adjustments at facilities thousands of miles away. That level of coordination requires the same semantic data foundations and organizational trust, just extended across every node in the network.

As one panelist observed, data is one of the few assets where the more it is used, the more valuable it becomes. The organizations that internalize that principle, treating data as something to be cultivated and utilized across every function, are the ones whose compounding data advantage puts them consistently ahead of their competitors.