Reading time: 9 minutes | Updated June 2026 | Applies to: Manufacturing · ERP · APS · OEE · Predictive maintenance · Digital transformation · Artificial intelligence
The misconception production managers most often hold about artificial intelligence is this: “AI is not our job, it is the software developers’ job.” Yet in a factory, the area where AI pays back fastest is not writing code but the ordinary decisions made every day: Which material should we order, and when? Does this batch carry a quality risk? Which machine will be the bottleneck next week? Today these decisions are mostly made on experience and intuition. Early-adopting manufacturers, however, have started to make the same decisions from data, and the gap widens a little more with every quarter.
Summary — The value of AI in manufacturing lies not in buying a new technology but in redesigning existing decisions. Early adopters are moving stock, quality, planning and maintenance decisions from intuition to data, and the competitive gap between them and those left behind is growing exponentially. In this article we look at how AI-supported decision-making in manufacturing actually works, what early adopters concretely gain, and why most projects stay stuck at the pilot stage. Within this frame we also explain the role of the solutions we at MindDX use to support these decisions with data: abas ERP, Odoo, SkyPlanner APS, Prodaso IIoT and ProcessMind.
Contents
- The misunderstood side of AI in manufacturing: decisions, not code
- What do early adopters gain?
- Where is AI-supported decision-making used in practice?
- Why do most AI projects stay at the pilot stage?
- The picture in Turkey: behind, but the window of opportunity is still open
- Conclusion: where to start?
The misunderstood side of AI in manufacturing: decisions, not code
The public perception of AI has largely been shaped by conversational language models. That is why many manufacturers see AI as a tool that “writes text” or “develops software”. The real value in manufacturing lies elsewhere: in systems that read signals from machine sensors, ERP records and process data and thereby speed up the decision process and improve the accuracy of decisions.
The critical point here is that AI does not replace people. McKinsey’s research shows that processes in which humans and AI work together are markedly more productive than those in which either works alone. The right approach is this: AI makes the options and the risk visible in advance; the domain expert makes the final decision. In other words, AI acts as an advisor here, not as the decision-maker. This approach is not only safer but also produces more accurate results, because the machine processes the data while the human handles the context and the exceptions.
What do early adopters gain?
AI is no longer a rare technology, but creating value from it is still rare. According to McKinsey’s late-2025 “State of AI” survey, almost nine in ten companies use AI regularly in at least one business function, yet most do not yet see a meaningful financial impact at enterprise level. The difference lies between those who own the technology and those who truly integrate it into their decisions.
Boston Consulting Group’s 2025 analysis sets out this difference clearly: only a small share of companies generate value from AI at scale, while the majority lags behind. According to BCG’s findings, companies that integrate AI into their processes achieve markedly higher revenue growth and productivity than competitors that do not. What matters is that this advantage is not a one-off: as early adopters accumulate data, their models improve and their decisions become ever more accurate, whereas companies that wait start from scratch with every new tool.
In manufacturing, two distinct profiles stand out: aggressive adopters, who quickly test new AI tools to optimise production and the supply chain, and the balanced majority, who wait until solutions mature and risks become clear. The second group preserves stability but misses out on speed gains and cumulative advantage.
Where is AI-supported decision-making used in practice?
To stay concrete, let us look at four areas in a production plant where AI today does not make the decision but supports it. What these four areas have in common is this: AI does not invent a new decision; it makes a decision that is already being taken earlier and more accurate.
1. The stock and purchasing decision“When and how much should we order?” is one of the costliest intuitive decisions in manufacturing. A wrong forecast leads either to a stock-out (production stops) or to excess stock (tied-up capital). According to McKinsey’s analysis of distribution operations, AI-supported demand forecasting can significantly reduce forecasting error and lower stock levels, while also substantially reducing sales lost to stock-outs.
The data foundation for this decision is a sound ERP system. abas ERP is designed to enable “fully informed decision-making” with KPIs, dashboards and business intelligence (BI) functions; for mid-sized and SME manufacturers, Odoo makes purchasing, stock and order processes visible in real time on a single platform. To understand at which stage these decisions get stuck, ProcessMind can determine from the ERP logs how the purchasing process (procure-to-pay) actually runs and at which step delays occur.
2. The quality decisionThe decision “Is this batch risky, should we ship it?” is traditionally made through sampling and manual inspection. AI-based machine vision and anomaly detection continuously analyse sensor data and can often show whether a batch is risky before production is even finished. BMW Group has been using AI-supported image recognition in series production since 2018; while detecting cracks, scratches and label defects, it also filters out “pseudo-defects”, thereby both safeguarding quality and freeing employees from repetitive inspection tasks.
The data source for this decision is the real-time signals coming from the machines. The Prodaso IIoT platform connects all machine tools, including older machines, and provides automatic anomaly detection and live quality indicators; according to MindDX data, installations can see scrap rates fall by as much as 15–30%.
3. The planning and scheduling decision“Which machine will be the bottleneck next week, which order should be processed first?” In most factories this decision is still made statically in Excel — while the situation on the shop floor is dynamic: machines break down, materials arrive late, urgent orders come in. SkyPlanner APS uses its Arcturus AI engine to continuously analyse machine availability, finite capacity, material readiness and order priorities; when a disruption occurs, it recalculates the plan within seconds. What matters is that it runs on top of the ERP/MES and does not replace the core systems — and a critical design principle: it does not replace the planner but provides a support layer, with the decision still coming from the planner.
The power of this can be seen in a concrete example from Turkey. The Izmit refinery of Tüpraş, which joined the World Economic Forum’s 2025 Global Lighthouse Network, raised its delivery reliability from 85% to 95% with AI-supported forecasting and optimisation solutions and significantly shortened its average loading time. This is the typical gain when planning happens in data rather than in Excel.
4. The maintenance decisionThe decision “Should this machine be stopped now?” is the most mature application area of AI in manufacturing. Predictive maintenance detects anomalies in sensor data and makes it possible to intervene before a failure occurs. According to McKinsey’s manufacturing analyses, predictive maintenance can extend equipment life, usually by reducing unplanned downtime significantly. BMW’s Regensburg plant prevents hundreds of minutes of production disruption per year in advance thanks to an AI system that uses existing sensor data. The foundation of this decision, too, is connected machines and a continuous flow of data; this is where Prodaso comes in with digital twins and predictive models.
Why do most AI projects stay at the pilot stage?
Honesty is needed here: AI projects frequently fail. According to RAND’s findings, the large majority of AI projects never reach production; a 2025 MIT study shows that the overwhelming majority of generative AI pilots deliver no measurable financial return. The root cause, however, is usually not the model or the algorithm, contrary to common belief, but two fundamental gaps:
The data foundation. AI can only provide decision support as good as the data it is fed. In a plant where the ERP system is fragmented, process data is invisible and machines are not connected, AI is not “smart” but “blind”. Industry research has repeatedly shown that the biggest cause of failure is data quality and readiness. Sound decision support therefore first requires a sound data infrastructure: connected machines (IIoT), an integrated ERP and visible processes.
People, process and the right start. In successful AI transformations, most of the effort goes not into the tool but into training people, updating the process and managing change. Another critical finding: companies that define a measurable success metric before starting the project succeed several times more often than those that do not. Research also shows that projects run with a specialist partner or started with ready-made solutions achieve far higher success than those that build everything in-house from scratch.
The picture in Turkey: behind, but the window of opportunity is still open
AI use among companies in Turkey is still low: according to TurkStat’s 2025 data, about 7.5% of enterprises use AI — a share that rises markedly among companies with more than 250 employees. The European Union average is around 20%. Although this picture looks like lagging behind, it actually opens a window of opportunity: because a large part of the market has not yet moved, the competitive advantage Turkish manufacturers who adopt AI early can gain is far greater today. The window of advantage is open — but it narrows a little more with every quarter.
Conclusion: where to start?
The answer to the question “When should we start using AI in manufacturing?” has become “the moment you start making your data visible”. Decision support does not begin as an all-encompassing project from start to finish; it begins with a single decision that causes you the most trouble — whichever of stock, quality, planning or maintenance is where you most often get the decision wrong.
When working with manufacturers in Turkey, the process we at MindDX follow is this: first build the data infrastructure (abas ERP or Odoo for integrated production, Prodaso IIoT for connected machines), then create visibility (seeing how processes actually run with ProcessMind), and finally speed up the decision process (basing planning on data with SkyPlanner APS). Decision support lasts when it rests on this foundation; imposed as a stand-alone tool, it stays at the pilot stage.
What early adopters gain is not mysterious at all: making the same decision earlier and more accurately than the competitor. If you know which decision in your production process rests “on intuition, not on data”, that is exactly the right place to start.
Let us identify together where AI-supported decision-making is possible in your production processes. Book a 30-minute assessment appointment based on your existing data in a free conversation with MindDX.
