Data and Knowledge: The Twin Foundations of Smart Manufacturing

In a training session on the factory floor, a veteran operator tells a new hire, “You’ll hear a vibration noise from the machine.”
“I’m not hearing anything,” the trainee replies.
“No, you’ll feel it,” the expert explains. “It’s an intuition I’ve developed working here for 30 years.”

That exchange, shared by Siva Lakshmanan, CEO of DeepHow, captures what's happening across manufacturing floors worldwide. On one hand, decades of operational expertise are walking out the door as experienced workers retire. On the other, mountains of production data sit trapped in disconnected systems, unusable for the decisions that matter most.

At Operations Calling 2025, industry leaders from DeepHow, Hexagon, and NVIDIA gathered to discuss how AI is transforming smart factories. The conversation centered on a critical gap: most manufacturers focus on either data infrastructure or knowledge capture, but rarely both. This article explores why AI only works when these foundations are built together, and how to get started.

Why Operational Knowledge Is Disappearing

The manufacturing skills gap isn't just about unfilled positions, it’s about the "brain drain" occurring as experienced workers depart. Industry projections estimate up to 30 million manufacturing jobs may go unfulfilled by 2030, but the deeper crisis is the loss of tacit knowledge: the decades of operational intuition that was never documented and is often impossible to articulate.

The Collapse of Natural Transfer

For decades, expertise was passed down through long-term apprenticeships. That model has collapsed due to two factors:

• Tenure: Average manufacturing tenure has dropped to under five years, meaning veterans leave before they can effectively train their successors.

• Complexity: Modern operations involve tighter tolerances and more sophisticated equipment, widening the gap between "following the SOP" and actually "doing it right."

Defining the "Know-How" Gap

"This knowledge is tacit in nature... it is not something that you can easily write up or even easily explain." - Siva Lakshmanan, CEO of DeepHow

Knowledge isn’t t found in a manual; it’s the operational intuition required to keep production flowing:

• Sensory Cues: Recognizing a vibration that signals trouble before a sensor triggers.

• Machine Nuance: Mastering the specific setup sequences that keep a temperamental machine running smoothly.

• Tribal Shortcuts: Knowing which error codes actually matter and which workarounds solve recurring problems.

For example, it can be compared to teaching a child to ride a bike: you can explain the physics, but the child must see it and feel it to understand.

"Knowledge is when it's sitting between your ears," he explained. "Know-how is when you're transferring it, and that's not an easy task at all." - Siva Lakshmanan, CEO of DeepHow

Without a systematic way to capture this "know-how," the expertise that took 30 years to build effectively disappears during a single exit interview.

Manufacturing floors generate a staggering variety of data, from PLC signals and time series to video feeds and sensor readings. However, this volume often remains untapped because of several core obstacles:

• The Silo Problem: Data is frequently trapped within specific departments, leading to ad-hoc collaboration rather than a unified strategy.

• The Value Gap: Having data is not the same as using it; value only arrives when data is converted into insights that inform other disciplines.

• The Context Crisis: Generic AI lacks the specific "know-how" of your local factory processes and machines.

• Garbage In, Garbage Out: High-speed GPUs and data centers are useless without high-quality, relevant data to feed them.

“We collect a tremendous amount of data whether through our software or through the devices that we put on the floor… it’s one thing to have the data and it’s another thing to be able to use that data.” — Hiren Kumbhojkar, Head of Product Management, Manufacturing Software Portfolio, Hexagon

The Solution: The goal is to move beyond simple collection toward a data strategy where information is structured, accessible, and contextually meaningful enough to drive real-time decision

The manufacturers making the most progress aren’t chasing big-bang transformation, they’re applying focused patterns that connect frontline knowledge with operational data in practical, scalable ways.

Here’s what that looks like:

Pattern 1: Capture What People Know, While They Work

The playbook:

• Record real work as it happens like setups, changeovers, troubleshooting, inspections

• Use AI to turn unstructured content (video, voice, gestures) into structured instructions

• Convert tacit know-how into searchable, repeatable, digital guidance

• Layer in checks, translations, and in-line support so new hires can apply it immediately

The result:

• Knowledge becomes visible and usable without pulling experts off the line

• Instead of one-time training, teams gain embedded guidance they can access anytime

• Expertise scales across shifts, roles, and locations with no documentation bottlenecks
  
  

Pattern 2: Turn Data Streams Into Real-Time Decision Support

The playbook:

• Fuse multimodal data like machine settings, camera feeds, quality logs all into one intelligence layer

• Train AI on your own historical data (programs, inspections, outcomes) to recognize patterns

• Deliver real-time guidance at the point of decision: “Here’s the fix,” not just “Here’s the problem”

• Use these insights to assist operators, not replace them and thus supporting human-in-the-loop workflows

“Insights have to be real time, right? Waiting two days for a report… that’s not tangible business value.” — Alvin Clark, Global Developer Relations Manager for Manufacturing and Industrial, NVIDIA

The result:

• Operators gain relevant, contextual support in the moment and not hours later

• Engineers get faster paths to root cause and stronger baselines for continuous improvement

• Your systems stop just reporting on performance and start helping improve it
  
  

The Common Thread

These approaches may start in different places, but they work for the same reason:

• They use local context, not generic models

• They make knowledge visible and data usable

• They embed intelligence where the work happens

• And they keep people in the loop and not out of it

That’s how manufacturers are making AI real, by connecting the foundations they already have.

You don’t need perfect data or complete knowledge capture to begin. Start with a real, recurring problem, then build both foundations in parallel. Most manufacturers follow a simple path:

• Assess your gaps in data accessibility and undocumented know-how

• Pick a high-impact use case like changeovers, troubleshooting, or training

• Capture knowledge and connect data as part of one focused workflow

• Give teams clear ownership and the guardrails to move fast. When people see their expertise reflected in the systems they use, adoption happens naturally, and improvement follows.

Tulip brings data and knowledge together in a single platform designed for the realities of frontline operations.

With Tulip AI Composer, teams can turn static SOPs and tribal knowledge into interactive, digital work instructions in minutes with no coding required. Upload a document, and the system extracts the steps, images, and logic needed to generate a working app. It’s a fast, repeatable way to make expert workflows accessible to every operator.

Pair that with AI Agent, Tulip’s embedded AI assistant, and operators get on-demand answers right in the flow of work. Whether they’re asking for torque specs, troubleshooting a machine, or reviewing historical performance, Copilot surfaces the right knowledge instantly pulling from your real SOPs, manuals, and production data.

Together, these tools create a closed loop: knowledge is captured, converted into applications, enriched with real-time data, and made accessible at the point of need. And because Tulip is part of a broader manufacturing ecosystem, that intelligence extends beyond a single platform. Through partnerships with innovators like DeepHow, Hexagon, and NVIDIA, manufacturers can connect expert knowledge capture, advanced simulation and digital twin capabilities, and accelerated AI infrastructure into the same operational workflow.

Tulip provides the foundation for a more connected, intelligent factory where AI supports both people and processes, and an open ecosystem allows that intelligence to scale across every operation.