Over the past three months, I visited 10 factories — food processing, metal hardware, chemicals, textiles. Workforces ranging from 50 to 2,000 people, locations from Guangdong to Jiangsu.
Before these visits, I expected to see cutting-edge examples of "smart manufacturing." After all, the buzzwords "digital factory" and "intelligent manufacturing" have been everywhere. The reality?
Of the 10 factories, 6 had installed so-called "smart systems," but fewer than 2 were actually using them.
The remaining 4 had spent money, installed systems, and ended up with wall-mounted display screens — props for visitors. What worried me more was that some of these systems weren't just useless — they were actively harmful.
If you're considering factory digitization or already on that journey, this article might save you a significant amount of money.
Let me share a real case. A food processing factory in Zhejiang, annual revenue around 80 million yuan. Last year, they spent 1.2 million yuan on a "digital workshop management system."
When I visited, there was an 86-inch display screen at the workshop entrance, flashing all sorts of data: production output, energy consumption, equipment status, quality rates... Impressive, at first glance.
But after chatting with the workshop director for half an hour, I discovered an awkward truth — the data on that screen was manually entered by him every morning. It wasn't automatically collected by the system. He walked around each workstation, checked the numbers, and typed them in.
I asked: "Besides displaying data, what else can this system do?"
He thought for a moment: "It can supposedly generate reports. But we've never used that feature. Excel works fine."
1.2 million yuan for a digital spreadsheet.
This wasn't an isolated case. Of the 10 factories I visited, at least 3 had similar situations — systems installed, data manually entered, screens serving as props for management tours.
This isn't intelligent manufacturing. This is "intelligent decoration."
If "intelligent decoration" merely wastes money, the following scenario is far more dangerous.
A hardware factory in Jiangsu installed an "AI scheduling system" last year. The vendor's proposal was impressive — AI algorithms would automatically optimize production scheduling, claiming a 15% capacity increase.
After one month of actual use, the workshop director went to the boss and said: "Can we turn this system off?"
Why? Because the production schedules the system generated had nothing to do with reality.
For example, the system scheduled Line A to produce 500 units of Product X that day. But the Line A operator knew that Product X's mold needed maintenance that afternoon, so production could only run until 2 PM. The system didn't know this. The mold maintenance schedule existed only in the operator's head, not in any system.
At 2 PM, Line A stopped. The system still showed "capacity progress: 60%." The next day, the system generated more unrealistic schedules. The operators stopped following the system and managed things themselves.
After one month, the system became shelfware. Worse — because it disrupted the original scheduling rhythm, that month's capacity actually dropped by 8%.
This isn't "useless" — this is "harmful."
I later spoke with the system vendor. They said: "The algorithm is fine. The client's data foundation was too poor, and the input parameters were inaccurate."
Translation: We sold a system that requires precise data to function, but the client had no precise data. Sell the system first, then ask the client to fix their data. The sequence was reversed.
After visiting these 10 factories, I identified several common characteristics of "fake intelligence." If you're implementing digitization, check your project against these:
The system requires workers to manually input data daily. This isn't intelligence — it's electronification. True intelligence means data is automatically collected and uploaded by equipment. Any step requiring human intervention is a design flaw.
The system's primary function is "showing data." There are dashboards, charts, and gauges, but no decision support, no automatic alerts, no closed-loop management. Data display is a means, not an end. If looking at data doesn't tell you what to do next, you've gained nothing.
The system's logic doesn't match shop-floor reality. Scheduling doesn't account for mold maintenance. Energy monitoring doesn't differentiate between shift-specific process variations. Quality systems can't correlate with equipment status. Workers simply don't trust the results.
Trying to do everything at once — scheduling optimization, quality traceability, energy management, equipment monitoring, warehouse logistics... and ending up doing nothing well. Intelligent manufacturing isn't about installing a complete system overnight. It requires phased, rhythmic progression.
After these 10 factory visits, I developed three core principles for understanding true intelligence.
Many factories follow this path: find a vendor → review proposals → buy a system → figure out how to use it.
The correct path is: identify problems → calculate value → find the minimum solution → implement → expand.
Every system that's actually being used started from a problem, without exception.
A textile factory installed a single current sensor on each loom, connected to a cloud platform for real-time equipment monitoring. No big screens, no scheduling optimization, no AI algorithms. But it solved one real problem: previously, when a loom broke down, operators discovered it only after two hours of downtime, losing thousands of yuan. Now, abnormal current triggers an alert within 5 minutes, and operators respond immediately, reducing downtime to under 15 minutes. Annual savings from reduced downtime: over 300,000 yuan. System cost: under 80,000 yuan. Simple, but effective.
I've seen too many factories try to make everything intelligent simultaneously. The result? Trying to grasp everything and grasping nothing.
The smartest approach: Solve the most painful problem first, demonstrate results, then move to the next.
Like building a house — you don't finish it in one day. Foundation first, then frame, then walls, then finishes. Factory intelligence works the same way: get equipment connected (foundation), then data collection and analysis (frame), then advanced functions like scheduling optimization and predictive maintenance (finishes).
But many factories skip straight to the "finishes" phase without laying a foundation.
What does "invisible" mean? It means frontline workers don't feel like they have an "extra system to operate." The system works silently in the background — alerting when needed, reminding when necessary.
If a system requires workers to spend 30 minutes daily entering data, it probably won't survive three months. Workers won't cooperate. Not because they're lazy, but because they're already busy enough.
A truly good system makes workers' jobs simpler, not more complicated.
If you're planning a factory digitization project, these recommendations may help:
Factory owners are bombarded with concepts — Industry 4.0, digital twins, AI scheduling, predictive maintenance... Each sounds impressive, each claims to solve your problems.
But few will tell you: What you probably need most right now is a current sensor on your air compressor.
Because that's not expensive, and it doesn't look "advanced" enough.
Doing less is harder than doing more. Especially in an industry where everyone is adding features.
But if you're a factory owner, what you really need may not be a "smart screen" to impress visitors, but a small system that can detect an air compressor anomaly at 3 AM and automatically alert you. The former makes you look impressive. The latter actually saves you money.
Which would you choose?
VoltKun specializes in AI-powered solutions for the energy equipment industry, helping manufacturers and distributors build truly effective intelligence systems.
Explore VoltKun Solutions →