Strategic imperatives for European manufacturers in China

Artificial intelligence (AI) today is less about AI as a human-like mind and more about creating and operationalising artificial knowledge: models that extract patterns from data and turn them into repeatable business outcomes. For European managers operating manufacturing plants in China, it is now imperative to integrate AI into their operations to gain a competitive edge, argues Gianluca Giorgi.

When people say AI, they often mean ChatGPT or another large-language model (LLM). However, this narrow view often misleads business leaders. AI is not an artificial mind; a more useful way of thinking about it today is as ‘artificial knowledge’ – systems that ingest data, learn patterns and produce actionable outputs. As we cannot fully define human intelligence, we should avoid attributing human abilities to these systems. Instead, we should focus on what they actually do: transform data into usable, repeatable knowledge that improves decision-making and operations.

AI predates recent startups: its roots reach back to early computing, expert systems—such as Industry 4.0 Expert, which can be used for process optimisation, predictive maintenance and intelligent decision-making—and decades of research. The modern renaissance—machine learning, deep learning and transformer architecture—was made possible by abundant data, more powerful hardware and improved mathematical techniques. For manufacturing, the critical change is availability. Sensors, enterprise resource planning (ERP) systems, manufacturing execution systems, and mobile connectivity have enabled factories to utilise masses of data. Where earlier automation projects stalled due to a lack of data or compute, today these inputs make practical AI use cases viable and profitable.

European managers and AI

European managers in China face a unique set of pressures: tight cost targets, customer expectations for higher quality and traceability, competition from both local and global players, and the need to align operations with company strategies. AI is not a magic bullet, but it is a force multiplier when applied right. The practical benefits are clear:

• Reduced unplanned downtime via predictive maintenance.
• Improved yield and quality through vision systems and process optimisation.
• Faster root-cause analysis by correlating sensor, production and quality data.
• Smarter workforce planning and a reduced administrative burden through automation.
• Improved lean systems with AI lean systems.

Common barriers to adoption in China

In the automation field and AI since the 1980s, three themes recur:

1. Many professionals remain attached to familiar routines and spreadsheet-based workflows, while managers often lack the time or inclination to explore advanced analytics platforms. Instead, they gravitate toward accessible tools like ChatGPT for translating documents or simplifying paperwork, rather than engaging with deeper system integration.

This pattern echoes the 1980s, when ledgers were processed both manually and digitally. At the time, computers were still viewed with scepticism, and many managers insisted on parallel records to ensure reliability.

2. Talent and skills gaps: China’s labour market has immense scale, but gaps exist in managerial AI literacy and systematic problem-solving approaches compared to Western best practices. Companies frequently struggle to recruit talent, but the faster route is to upskill current managers.

3. Data and system fragmentation: Legacy ERPs, Excel sheets, and stove-piped systems yield poor data hygiene. Without trustworthy data, large AI projects will fail.

In order to scale up AI pilots, European managers need a grounded, staged approach. These four pillars are essential to the success of AI projects: use cases, data and architecture, people and skills, governance and measurement.

Use cases: Start small

• Choose high-impact, low-complexity pilots: predictive maintenance on a critical machine line, defect detection with computer vision, or automated key performance indicator (KPI) dashboards that combine production, quality and maintenance data.
• Define measurable objectives, such as reductions in downtime, increases in yield, or hours saved in manual reporting.
• Time-box pilots: Three to six months with clear success criteria and a go/no-go decision.

Data and architecture: Pragmatism wins

• Tidy the inputs: Start with a small, trusted dataset rather than an ambitious ‘data lake’ approach. Clean, labelled data beats bigger but noisy datasets.
• Hybrid architecture: Use edge computing for real-time inference (e.g., vision systems, programmable logic control data) and cloud computing for model training and heavy analytics. Smaller, purpose-built models often suffice on the edge and reduce data transfers and costs.
• Leverage modern, pragmatic tools: Low-code/no-code workflows for data ingestion, model monitoring for drift, and machine learning operations practices to ensure reproducibility and version control.

People and skills: Train managers to be promoters and sponsors

• Focus on managers as AI enablers, not just consumers. A European manager in China must become an expert at using AI applications.
• Upskill existing talent: Create targeted programmes that teach plant managers how to interpret model outputs, design experiments and manage change. It is not necessary to have people with advanced education working on every line – experienced operators and engineers make excellent ‘AI translators’ when trained.
• Embed learning into daily work: Design learning opportunities that are part of routines, e.g., retrospectives tied to model-driven interventions or short ‘show-and-tell’ sessions for shifts.

Governance and measurement: Keep it practical

• Establish a lightweight governance model: Data quality owners, model stewards for each use case, and an escalation path for ethical or safety concerns.
• Measure return on investment (ROI): Connect model outputs to operational KPIs and financial impact. If a model does not demonstrably reduce scrap, downtime or cost per unit, stop or pivot the project.
• Protect security and intellectual property (IP): Industrial data is sensitive. Ensure encryption during transmission, maintain robust access controls, and have clear policies for model and data sharing.

Technology choices: Size and placement matter

There is a tendency to equate bigger with better. In manufacturing, smaller, focussed models often provide the best ROI. Multiple small models can be orchestrated to address discrete tasks (anomaly detection, scheduling, quality classification) and can run on edge devices for low latency and resilience. For language tasks (documentation, standard operating procedure translation), leveraging LLMs via secure private endpoints or fine-tuned smaller open models often provides the most practical balance of performance, cost and data privacy.

A cautionary note on LLMs and ‘AI everywhere’

Tools like ChatGPT are useful for documentation, translation and ideation – but they are not a substitute for production-grade AI systems that manage real-time operations. Overreliance on general-purpose models without domain adaptation or guardrails can create errors and false confidence. Think of ChatGPT as a rapid prototyping and communication assistant, not a direct replacement for calibrated models embedded in the control loop.

Change management in practice

Ever since personal computers began to automate many workplace processes, the hardest obstacle has never been technology – instead, it has been convincing people to change their daily work and to trust new processes. In practice, introducing AI applications often brings early resistance from both managers and operators. It is important to demonstrate small wins quickly and publicly, such as reduced rework on a single product line, or a weekly maintenance alert that prevents a costly breakdown. It is best to identify ‘change agents’ – generally the most respected operators and managers who can champion the new solutions and provide peer-to-peer training.

Developing internal talent

Many European plants in China report difficulty hiring staff with the required skills. The fastest scalable strategy is to develop internal talent:

• Train managers in applied AI literacy – how to scope projects, read model outputs and ask the right questions.
• Turn experienced operators into ‘AI operators’ by pairing them with data engineers and giving them practical training on data labelling, validating model outputs and continuous improvement.
• Create career pathways that reward cross-disciplinary skills: production expertise plus data-enabled decision-making.

Measuring success: Metrics that matter

It is essential to move beyond ‘vanity metrics’. It is best to track KPIs directly tied to business outcomes, for example:

• Operational: Meantime between failures, overall equipment effectiveness, yield, cycle time.
• Financial: Cost per unit, scrap and rework costs, maintenance spend.
• People: Time saved on reporting, number of employees trained, adoption rates of model-driven processes.

Risks and mitigation

• Model drift and false positives: Run continuous monitoring and retrain models when data patterns change.
• Data privacy and IP leakage: Segregate production data, use on-premise inference for sensitive workloads, and set strict vendor contracts.
• Energy and compute costs: Favour efficient models, schedule heavy training tasks at off-peak times or use cloud computing space, and use edge inference for real-time tasks.

AI is not a futuristic intelligence that replaces human leadership. It is a set of techniques that codify experience into repeatable, testable knowledge. For European general managers in China, the strategic imperative is simple: treat AI as an operational technology that requires the same discipline as any physical asset – clear objectives, measured outcomes, trained people and robust governance. It is essential to start small, focus on measurable wins, train managers and operators, and scale with a pragmatic architecture. The prize is significant: better quality, lower costs, faster decision making and stronger resilience in a highly competitive manufacturing environment.

Gianluca Giorgi is the CEO of ES Automation Consulting. He has worked in automation and AI since the 1980s, leading early industrial AI projects. He is a board member of the European Chamber and holds diplomas from MIT in Blockchain (2021) and AI Data Science (2024).