A digital twin sounds simple: a virtual replica of a physical thing. In practice, building one that actually works requires better data infrastructure than most organizations have.
GE started talking about digital twins for jet engines about a decade before the concept became a buzzword. The idea was straightforward in principle: instrument an aircraft engine with sensors, stream that sensor data to a model of the engine, and use the model to predict when components would need maintenance before they actually failed. Instead of replacing parts on a fixed schedule (some components replaced too early, some replaced too late), you could replace them based on actual condition. The maintenance savings were real. The safety benefits were real. The concept worked because GE had decades of engineering data on exactly how those engines behaved, the sensors were purpose-built and carefully calibrated, and the engineers interpreting the model's outputs understood both the physical system and its digital representation.
That specific set of conditions is what gets left out when people talk about "just building a digital twin" for a factory floor, a supply chain, a hospital, or a smart city.
A digital twin, in the cleanest definition, is a real-time virtual replica of a physical thing, process, or system that stays synchronized with its physical counterpart through live data. The synchronization is what distinguishes a digital twin from a simulation. A simulation models how something might behave under certain conditions. A digital twin reflects what is actually happening right now, updated continuously as sensor data arrives. You can run scenarios against it (what happens if this machine slows down?) but the baseline is always the current state of the real thing.
Gartner has placed digital twins on its Hype Cycle multiple times across different categories, which you can see in their methodology page at https://www.gartner.com/en/research/methodologies/gartner-hype-cycle. Gartner's Hype Cycle framework plots technologies across five phases: Innovation Trigger, Peak of Inflated Expectations, Trough of Disillusionment, Slope of Enlightenment, and Plateau of Productivity. Digital twins as a concept have gone through an interesting trajectory on that cycle, appearing in multiple domain-specific variants (manufacturing, smart cities, IoT) at different stages simultaneously. I am hedging specific placement claims here because the Hype Cycle updates annually and the exact positioning at any given moment depends on which specific variant you are looking at.
Siemens, PTC, and Dassault Systèmes are the names you see most often in industrial digital twin deployments. These are companies with deep roots in product lifecycle management (PLM) and industrial simulation software. Their digital twin offerings extend their existing simulation platforms with live data connectivity. Siemens calls their industrial platform Xcelerator. PTC built its digital twin capabilities on top of its ThingWorx IoT platform. Dassault has the 3DEXPERIENCE platform. These are not lightweight tools: they are enterprise-grade systems with corresponding enterprise-grade implementation complexity and cost.
What strikes me as an IS researcher is that a digital twin is fundamentally a sociotechnical artifact, not just a technical one. The data pipelines that feed it are sociotechnical: they require decisions about which sensors to deploy, how to calibrate them, how frequently to sample, how to handle missing or anomalous readings, and who is responsible for maintaining the data quality over time. The model at the center of the twin is sociotechnical: it encodes assumptions about how the physical system behaves, and those assumptions were made by engineers who may have left the organization, under operating conditions that may have changed. The outputs of the twin are sociotechnical: they land in the hands of operators and managers who have to decide what to do with them, and whose trust in or skepticism toward the system shapes whether the twin's recommendations are actually acted on.
The data infrastructure challenge is where most enterprise digital twin projects run into trouble. Building a useful digital twin requires knowing the real-time state of the physical system it models. That requires sensor coverage, network connectivity, data ingestion pipelines, and data storage with low enough latency to be practically "real-time" for the use case at hand. Many industrial facilities have partial sensor coverage from sensors installed at different times by different vendors using different protocols. Connecting those data sources into a coherent, consistent stream is not a trivial integration problem. MQTT, OPC-UA, Modbus, and proprietary machine protocols all coexist in the same factory, and getting them to talk to each other in a format the twin's model can use is engineering work that the marketing materials for digital twin platforms tend to underplay.
Supply chain digital twins have attracted attention as a concept since the COVID-19 pandemic exposed how brittle global supply chains were. The idea of having a real-time model of your entire supply chain that you could query to find out where disruptions were propagating and what your options were for rerouting: that was appealing to a lot of operations leaders who had just spent two years dealing with shortages they could not see coming. The reality is that a supply chain digital twin requires data from dozens or hundreds of suppliers, logistics providers, and ports, most of whom do not have sensors or APIs or any interest in sharing their operational data with you. The organizational and contractual problems dwarf the technical ones.
Smart city digital twins, the most ambitious variant, add a political dimension. A real-time model of a city requires data about traffic, utilities, buildings, and people. The people dimension has obvious implications for surveillance, data governance, and civic consent. Several cities have announced digital twin initiatives. How many have meaningfully deployed them, and under what constraints, is a question where I would want to read more recent reporting before drawing strong conclusions.
The technology for digital twins is real and improving. The gap between the concept and its useful deployment in most organizations is not a technology gap. It is a data quality gap, a data governance gap, and an organizational readiness gap. GE's jet engine twins worked because GE had decades of investment in exactly the right foundational conditions. What most enterprises actually have is a long way from that.
About the author
Share
More notes
Related notes