"Digital twins are a great concept, but very few of them actually work on the shop floor."
Digital twins have 🙋 been running as manufacturing's "next paradigm" for nearly a decade since the mid-2010s, with adoption decisions numbering in the hundreds of thousands. Unfortunately, however, the cases that have proven business value in actual operations remain at a single-digit level.
According to a 2025 McKinsey survey, 71% of digital twin PoCs end at the "visualization level," and only 9% consistently contribute to operational decision-making. Where is the line that separates hype from reality?
The value of a digital twin comes not from the "3D model" but from the "decision-making loop."
1. The Four Maturity Stages of Digital Twins
A digital twin is not a single technology but a continuum of maturity. The stages fall broadly into four levels. ① Visualization twin — a 3D model plus a real-time data overlay. ② Analytics twin — cause-and-effect analysis based on historical data. ③ Predictive twin — forecasting future scenarios through dynamic simulation. ④ Autonomous twin — the twin feeds back directly into the system and determines the control loop.
Most digital twin projects stop at Stage 1, the visualization level. The problem is that business value only truly materializes at Stage 3 and above.
- Stage 1: Visualization — a moving 3D model (a pretty dashboard)
- Stage 2: Analytics — deriving cause and effect from dynamic data
- Stage 3: Prediction — physics-and-data hybrid simulation
- Stage 4: Autonomy — bidirectional control between the twin and the physical system
2. The Patterns of Failing Digital Twins
The common threads in failure cases are clear. First, they start out disconnected from operational KPIs. The "let's build the twin first" approach locks the project into the visualization stage.
Second, data quality is not ready to support the twin. Even a simulation that is theoretically accurate cannot properly guide operational decision-making when its input data is flawed. Third, the effort ends as a one-off project and never carries over into an operational pipeline.
A digital twin is not a "project" but an "operating framework."
3. What Successful Digital Twins Have in Common
The 9% of success cases share common conditions.
- Achieved a direct link to operational KPIs within one year: 87% of successful projects
- Completed data standardization first: 92%
- Reached Stage 3 (prediction) or higher: 78%
- Twin embedded in an actual control loop: 65%
- Average ROI payback period: 14 months
4. Case Study — Adopting an Autonomous Twin in Shipbuilding
A Korean shipyard ran a simple visualization dashboard (Stage 1) for two years to reduce the quality defect rate in its welding process, but the defect rate did not change. Root-cause analysis showed that although visualization was in place, the causal relationships between working conditions and weld defect data were not being fed into the operational pipeline.
After building out data standardization plus an analytics twin (Stage 2) and a predictive twin (Stage 3) step by step on PlantPulse, the shipyard cut its weld defect rate by 31% within nine months and reduced rework costs by roughly KRW 1.7 billion per year. What mattered was not the digital twin itself, but its connection to the operational pipeline.
How PlantPulse Answers This
PlantPulse designs the digital twin not as a "standalone project" but as "part of the industrial data operations pipeline." The ISA-95 asset model, time-series data, events and alerts, and predictive models are unified into a single layer, so the twin's outputs flow naturally into operational workflows.
So that the twin does not stop at "showing" but becomes something that "decides and controls," governance, control permissions, and audit trails are built into the twin architecture as well.
Closing Thoughts
What separates digital twin hype from reality is not the level of technology but the "connection to operations." A twin that stops at visualization is merely a cost center; only a twin connected to operations proves its value.
If you are considering a digital twin, first answer this question: "Which decisions in our operations will the twin replace?" If that answer is not clear, the outcome will end as hype.
© KOPENS — Industrial DataOps & PlantPulse Platform