Artificial intelligence has become one of the most discussed topics in boardrooms across the manufacturing sector. Vendors promise smarter factories, lower costs and better decisions. Yet many companies that rush to buy AI tools discover an uncomfortable truth: the algorithms are only as good as the data they are given, and much of that data is not ready.
The data problem
In many factories, operational information is scattered across systems that were never designed to talk to each other. Production figures sit in spreadsheets, machine data stays locked inside equipment controllers, quality records are kept on paper forms, and inventory lives in a separate accounting system. Even when data exists, it is often incomplete, recorded in inconsistent formats, or missing the timestamps needed to link one event to another.
An AI model trained on this kind of information will produce unreliable results. A forecasting tool cannot predict demand accurately if past sales and production records do not match. A quality model cannot learn what causes defects if defect reports are not tied to the machine, shift and raw material batch involved.
Practical first steps
Getting data in order does not have to be a massive project. Companies that succeed with AI usually begin with a few basic disciplines:
- Start with a business question. Decide what problem matters most, such as reducing scrap, cutting energy use or improving delivery times, and identify which data is needed to address it.
- Digitise key records. Replace paper logs for production, downtime and quality with simple digital forms or tablets on the shop floor.
- Connect the machines that matter. Collect data from the most critical equipment first rather than trying to connect everything at once.
- Agree on definitions. Make sure terms like "downtime", "defect" or "output" mean the same thing across departments and shifts.
- Assign ownership. Someone needs to be responsible for keeping data accurate and complete over time.
People matter as much as technology
Data readiness is also a question of skills and culture. Operators who record information need to understand why it matters, otherwise entries become careless. Managers need enough data literacy to ask good questions and to interpret the answers an AI system provides. Many firms find that training existing staff, who already understand the production process, is more effective than relying solely on outside specialists.
Security deserves attention as well. Connecting factory equipment to networks and cloud services creates new entry points for cyberattacks, so access controls and network segmentation should be planned from the start rather than added later.
An opportunity for smaller manufacturers
For small and medium-sized industries, which form a large part of Indonesia's manufacturing base, this message is encouraging. Getting value from AI does not necessarily require a large upfront investment in advanced software. Cloud-based tools have made analytics far more affordable than a decade ago, but they only deliver when they are fed with clean, consistent information.
The companies that will benefit most from AI in the coming years may not be those that buy the most sophisticated tools first. They are more likely to be the ones that quietly build a solid foundation of reliable data, so that when they do adopt AI, it has something meaningful to learn from.
