AI in manufacturing: Why data quality determines the success of artificial intelligence

Artificial intelligence (AI) is now considered a key technology of Industry 4.0.

However, practical experience shows that only clean, semantically consistent data creates the conditions for AI applications to generate real added value in factories.

In a presentation at the Industry Forward Expo, Dr. Ulrich Ochs, Managing Director of FORCAM ENISCO, explained why data quality is the central prerequisite for the successful use of AI apps – and how companies can master the complexity of production processes. You can watch the presentation here as a video. Important findings and recommendations:

Artificial intelligence: The more complex the processes and systems, the more important data quality becomes

Artificial intelligence has long been used in many areas—from voice-based AI applications such as ChatGPT, Perplexity, and Mistral to image processing and assistance systems in machine operation.

However, different rules apply in production than in traditional Internet services.

Dr. Ochs aptly describes this:
“AI is not a panacea that can solve all our problems on its own.”

Simpler AI models can now easily stabilize physical states or recognize patterns in data. But as soon as production facilities, machines, sensors, IoT structures, and entire factories come into play, the complexity increases massively.

The rule of thumb is: the more complex the data models, processes, and systems, the more difficult it is for AI to deliver consistent results – especially in real-time operation.

AI tools and AI applications in use: What really matters

How should companies proceed if they want to use AI-supported applications in their production processes?

Dr. Ochs emphasizes that manufacturing teams should first define the desired output:
“You have to think in terms of your goal: What do I want to achieve with AI? Then you can work backwards to figure out what data you need to do that.”

AI technologies can only process what is available in their input data in a targeted and “intelligent” manner. Systematic data management enables scalable AI solutions in industrial operations.

Anomaly detection in production—a frequently requested use case

One of the most common industrial use cases is anomaly detection.

A sensor provides measured values from the process. However, latencies, i.e., time distortions, occur as the data passes through networks, protocols, transformation steps, and IT systems. This alters the originally clean signal.

Dr. Ochs explains why this is critical:
“AI has a problem recognizing what is due to latency in signal distortions and what is a real error in the signal.”

The solution: signal acquisition as close to the machine as possible.

AI-supported voice dialogues: A manufacturing chat as a new area of application

Another example is an AI-supported manufacturing chat—a language model that works like ChatGPT but specifically accesses production data.

The benefits:

• Employees can ask questions verbally using speech recognition
• AI systems answer these questions in real time
• Diagrams, trends, or status information are generated automatically
• Failures and malfunctions can be analyzed more quickly

But here, too, the same rule applies: without consistent data, there can be no reliable answers.

Truly intelligent AI requires semantics: uniform data models are crucial

If one machine classifies good parts as “yield,” but another uses the text “good part,” AI does not automatically link the two different terms to a value.

If such classifications are not clearly defined, this prevents the optimization of production processes, complicates analysis, and blocks the use of AI technologies across multiple systems.

Only when all machines and IT systems use the same semantics can AI systems unfold their full potential.

Deep learning, IoT, and cloud—but with structure, please

Modern industrial companies rely on cloud services, IoT platforms, and deep learning algorithms.

But without uniform data models, the added value is lost.

Dr. Ochs sums it up:
“It is important to make data smart: assign meaning to it as early as possible and distribute it to all IT systems as a single source of truth.”
This makes artificial intelligence an integral part of all business processes—from production and maintenance to quality assurance.

AI solutions in industry: Rethinking organizational structures

Many companies are wondering whether they need to create a new role, such as a data manager.

Dr. Ochs comments:
“The introduction and use of AI is neither a purely IT nor a purely OT task. Both areas—IT and OT—must work closely together.”

The development toward AI-supported factories is therefore not only a technological challenge, but also an organizational one.

Four recommendations for AI in manufacturing

• Think from the end

• Stay lean

• Make data smart

• Close data gaps

Conclusion:

AI in manufacturing requires clean data, clear structuring, and consistent semantics.

Artificial intelligence offers enormous potential for increasing efficiency, automation, quality, and process optimization in industrial companies. However, its success depends crucially on the correct collection of data, its semantic meaning, and cross-system usability.
With clean data, the simplest possible structure, a clear semantic model, and a cross-functional team tailored to the size of the company, AI can create new added value such as anomaly detection or smart manufacturing chats. Then it has what it takes to become a key technology in a modern manufacturing infrastructure.