A wave of capital investment in physical infrastructure is driving new facility builds and making consistency at scale and cost control more critical than ever.
Beyond the shortage of skilled IT engineers across industrial companies, many organizations still face a basic question: where to start. Factories with multiple generations of equipment were not designed for a digital environment. As a result, the industry needs a framework that enforces engineering rigor and operational discipline.
The Infrastructure Reality Check
Before investing in IoT for manufacturing, the priority is to clearly define whether the objective is scaling operations or maintaining competitive positioning. Multi-site operations are increasingly adopting standardized processes to ensure consistent production across locations.
In some cases, the goal is to establish engineering rigor through validated processes, ensuring the organization can replicate these processes across sites as it adopts an AI-driven manufacturing approach. In others, the focus is purely economic: even a small number of sensors paired with predictive analytics can prevent breakdowns and reduce downtime, saving millions.
To avoid disrupting operations, digital transformation should be rolled out in phases, with each phase addressing lead time reduction and inventory carrying costs. After assessing the current technical and digital landscape, the next step is to evaluate solutions against engineering rigor and cost control.
Once the technology path is clear, the next step is legacy modernization.
Connecting Legacy Assets
Until recently, legacy systems on the factory floor were a major barrier to change. The challenge was not just outdated languages but also critical business logic buried inside black-box systems.
Today, AI can support anomaly detection and help teams take control of legacy systems. For instance, in a recent project with a mid-sized manufacturer, we assessed ERP, MES, SCADA and PLC systems in a short timeframe, while automatically documenting the legacy code. Data was mapped into standardized production data models.
We integrated through existing layers like OPC UA and middleware, without touching control logic, and streamed data into a shared event model that separated AI from legacy systems.
Bridging the IT/OT Divide
No one shuts down a million-dollar machine just because it does not support modern protocols, and no one replaces a system built on something like Windows NT 4.0 overnight just for better features. In manufacturing, the cost of downtime is simply too high.
From Physical Sensors to the Digital Layer
To enable continuous monitoring of legacy equipment, machines first need a way to connect to digital systems. When built-in connectivity is missing, the most practical approach is retrofitting with external smart devices that capture physical signals from key components.
Capturing machine signals through IIoT is only the starting point. To make this data usable, it must align with standardized production data models like OPC UA, historians and middleware.
The data is then processed in a unified pipeline: signals are cleaned, formats are standardized, timestamps are aligned and linked to lot/serial traceability and batch traceability data. This often happens at the edge, before the data is sent through gateways into a shared event model. Once structured, each asset can publish its state via protocols like MQTT.
Making sense of the data (UNS)
One of the core challenges is the planning vs execution gap caused by fragmented data. The only scalable solution is to manage inventory as a real-time operational discipline, replacing scattered inputs with a modern Human-Machine Interface (HMI) for production and workflow monitoring. This leads to a single source of truth built on a Unified Namespace (UNS).
In traditional setups, systems are connected point-to-point: MES to ERP, SCADA to BI, plus AI on top. These tight links create dependencies, where any change in one system affects others. With a UNS approach, systems do not integrate directly. Instead, they publish and subscribe to events in a shared data layer.
In one project, a legacy MES required core changes for every new integration, slowing releases to weeks or months. Moving to an event-driven model changed that. Assets began publishing their state via MQTT to the UNS, while new logic ran in separate containerized services subscribing to those events. This decoupled architecture allows teams to launch new features independently, cutting delivery cycles from weeks to days.
At its core, a UNS brings all factory data into a single real-time view. It gives clear visibility into operations at any required frequency, from hourly to sub-second updates, and creates the foundation for the next step: IIoT enables production and workflow monitoring, while AI provides exception handling.
AI as an Operational Discipline
For manufacturers, building AI models from scratch rarely makes business sense. It is expensive, slow and better suited to large R&D programs. A more practical approach is to use proven components that already understand signals or visual data and combine them to fit specific shop-floor needs.
Once the relevant AI capabilities are clear, the question becomes how to deploy them without heavy in-house research. In most cases, vendors build custom solutions on top of existing models and train them on production data. Combining computer vision with neural networks can create high-value, production-specific assets that are difficult to replicate because they depend on unique operating conditions.
At the same time, AI depends on data quality. It does not determine truth; it reflects the inputs it receives. That makes data quality, pipelines and storage design critical and requires strong data engineering. Crucially, maintaining release discipline ensures that digital systems do not accumulate complexity that compromises operational reliability.
The priorities are lead time reduction and cross-site consistency and integrating innovation without disrupting existing operations. Done right, digitizing even long-established processes does not compromise stability and creates a clear competitive edge: those who achieve consistency at scale win and those who treat inventory as a live operational discipline will maintain a capital-efficient edge in any market.
Kostiantyn Gitko, CEO, Devox SoftwareDevox Software
Kostiantyn Gitko is the CEO of Devox Software.
