AI in Manufacturing: Where It Delivers Value Today

AI in manufacturing: between the promise and the shop floor

Few topics in manufacturing are discussed as intensely – and concretely as rarely – as artificial intelligence. In keynotes and pitches, AI sounds like self-optimizing factories. On the shop floor, the real question is simpler: which of this actually helps today, on a real line, with the machines and data we already have?

The short answer: AI manufacturing software delivers measurable value where it meets clean, contextualized production data. Where data sits in silos – machines here, orders there, quality in yet another system – even the best model stays ineffective. This article shows where the value is real today, the one prerequisite that decides it, and what can actually be proven in practice.

Why most AI projects fail on data, not on the model

Manufacturers rarely lack data. Machines emit signals, the ERP holds orders, quality assurance documents deviations. What is missing is the operational context: these sources don't speak the same language and are seldom linked in real time.

Yet an AI meant to assess a downtime event needs exactly that context – which machine, which order, which material, which shift, which preceding events. Without a shared data model, the model computes on fragments and produces results no one trusts. The first step toward usable AI in production is therefore not AI at all – it's a consistent data foundation.

Where AI in manufacturing already helps today

Beyond the future scenarios, there are use cases running in production right now. At SYMESTIC, three AI features are available directly inside the cloud-native MES platform:

• Automatically translating and classifying downtime reasons. AI maps recorded downtime consistently to the right categories – instead of inconsistent free text, you get an analyzable downtime history that lean and maintenance teams can use directly.
• AI-assisted configuration instead of manual regex work. Defining patterns and parsing rules for machine data is classically error-prone expert work. AI-assisted configuration lowers that barrier and speeds up machine onboarding.
• Shift-log summaries. From a shift's entries, the AI produces a concise summary – handovers get faster and nothing relevant gets lost in the detail.

These features share one principle: they remove repetitive, manual analysis and documentation work and make existing data usable faster – without anyone having to launch an AI project.

The prerequisite: a single data model

For AI in manufacturing to work reliably, machines, orders, quality and tribal knowledge have to converge on one data model. SYMESTIC uses an ISA-95-based data model: machine signals (via OPC UA or an IoT box, from a 1990s controller to a current machine), ERP orders and quality data are contextualized in real time and made comparable across plants.

Because the platform is cloud-native on Microsoft Azure, this foundation comes without your own server or IT project. That is the difference between "we have data" and "we can work with our data" – and the real starting condition for any AI application in the plant.

AI needs trust: evidence-backed and human-in-the-loop

In manufacturing – and especially in regulated industries – an AI answer is only worth something if it is traceable. Three guardrails are decisive:

• Evidence-backed: every AI statement points to the underlying production data instead of "guessing".
• Human-in-the-loop: writing or intervening actions are confirmed by people, not executed autonomously.
• Read before write: AI first analyzes and suggests; interventions in production follow only in a controlled, safeguarded step.

This approach is what separates an impressive demo from a feature that actually gets used in shift operations.

What it delivers in practice

The value of a contextualized data foundation shows up in concrete customer results. At Meleghy Automotive, SYMESTIC was rolled out across five plants in four countries in six months – integrated with SAP R/3. The result: roughly 10% fewer unplanned downtimes and about 7% higher output. At Carcoustics, an existing MES with more than 500 machines was replaced within six months.

Customer Story Automotive

How Meleghy Automotive connects its plants with a single MES platform

• Standardized MES platform across all sites
• Real-time transparency in body production
• OEE & loss reasons captured automatically
• Scales from pilot to series production

Read case study Approx. 4 min read · free

Rollout Multi-site MES

Case Study

Across use cases, the typical effects of end-to-end real-time transparency fall into these ranges: +5–10% OEE, up to +15% productivity, −5–10% energy consumption and −5–15% scrap. AI here is not an end in itself – it's the lever that turns this data foundation into decisions faster.

Outlook: AI agents and the factory as a data source

The next stage is already in development – and we label it deliberately as planned, not available today:

• MCP server (planned, Q4 2026): production data such as OEE, downtimes and order status become readable for external AI assistants – e.g. Claude, ChatGPT, Gemini or Microsoft 365 Copilot. Plant knowledge becomes semantically searchable.
• AI write actions (planned, 2027): anomaly detection and controlled, human-confirmed interventions directly from the system.

The direction is clear: away from AI as an isolated feature, toward a production platform whose data is accessible to people and AI agents.

Conclusion

AI in manufacturing is no longer just a future promise – but it isn't automatic either. The value doesn't come from the model; it comes from the data foundation. Only when machines, orders and quality share one real-time context does AI produce results you trust in shift operations. So if you're starting today, don't begin with the AI tool – begin with the question of whether your production data converges in the first place.

About the author

Mark Kobbert

CTO at SYMESTIC. 12+ years building the cloud-native MES platform on Microsoft Azure — microservice architecture, IoT gateway development, real-time data processing for 15,000+ connected machines across 18 countries on four continents. B.Sc. Wirtschaftsinformatik (SRH Hochschule Heidelberg). Expertise: cloud-native MES architecture, Microsoft Azure, microservices, OPC UA, MQTT, IoT gateway development, edge computing, ISA-95 integration, ERP-MES integration, brownfield machine connectivity, real-time data processing, IT/OT convergence. · LinkedIn