Digital twins have become one of the most enduring ideas in modern infrastructure technology. From smart cities to power grids, from transport networks to industrial plants, the promise is compelling: a living digital replica of physical infrastructure that can predict failures, optimise performance, and guide decisions before they are executed in the real world.
In presentations and pilots, digital twins look transformative. In live infrastructure, they routinely disappoint. This gap is not a tooling problem, a data problem, or a maturity problem. It is a systems problem.
The core reason digital twins struggle outside controlled environments is simple but uncomfortable:
Operational entropy in real infrastructure exceeds the assumptions embedded in most digital twin models.
Until this mismatch is understood, digital twins will continue to cycle through hype, pilots, partial deployments, and quiet abandonment—only to re-emerge under a new label a few years later.
The Original Promise of Digital Twins
At their conceptual best, digital twins aim to do three things:
1. Represent reality faithfully
A digital model mirrors the physical asset, process, or system with sufficient fidelity.
2. Stay synchronised with the real world
Sensor data, telemetry, and operational updates keep the model current.
3. Enable safe experimentation
Scenarios can be simulated digitally before acting physically.
In tightly controlled environments—manufacturing lines, aerospace systems, or single-owner industrial assets—this promise often holds. These environments share common traits:
- Clear system boundaries
- Stable operating conditions
- Centralised control
- Enforced process discipline
Infrastructure systems share none of these characteristics.
Infrastructure Is Not a Closed System
Digital twins implicitly assume that the system being modelled is bounded. Infrastructure is not. Live infrastructure operates as an open, socio-technical system, shaped by forces that resist precise modelling:
- Human behaviour
- Institutional decisions
- Regulatory interventions
- Weather and climate variability
- Political and economic shocks
- Maintenance shortcuts and workarounds
These factors are not edge cases. They are the system.
A digital twin can model pipes, transformers, trains, or substations. What it cannot reliably model is:
- How operators actually respond under pressure?
- How procedures drift over time?
- How incentives distort decisions?
- How rules are interpreted in practice?
The moment a digital twin leaves the lab and enters live operations, it encounters operational entropy.
Understanding Operational Entropy
Operational entropy is the accumulation of disorder, variability, and unpredictability in real-world systems over time. In infrastructure, entropy arises from:
- Incremental changes made without architectural oversight
- Temporary fixes that become permanent
- Staff turnover and skill erosion
- Partial compliance with procedures
- Divergence between documented and actual workflows
Digital twins tend to assume:
- Consistent inputs
- Stable configurations
- Rational decision-making
- Clean data flows
Live infrastructure delivers:
- Incomplete data
- Asymmetric information
- Delayed feedback
- Context-driven decisions
The result is not a small modelling error. It is a structural divergence between model and reality.
Why Digital Twins Work in Pilots?
Digital twin pilots usually succeed for the same reason many InfraTech pilots succeed: they operate in artificially low-entropy conditions.
Pilots are typically:
- Narrowly scoped
- Heavily supervised
- Staffed by motivated teams
- Shielded from institutional friction
Data quality is curated. Processes are followed. Exceptions are minimised.
In effect, the pilot environment is shaped to fit the model.
Scaling reverses this relationship. The model must now fit the environment. That is where failure begins.
The Scaling Cliff
When digital twins move from pilot to production, three things happen simultaneously:
1. System Boundaries Expand
The twin must account for interactions with adjacent systems—many of which were never designed to be modelled together.
2. Control Dilutes
Decisions are no longer made by a small expert group, but by distributed actors with different priorities and risk tolerances.
3. Entropy Accelerates
Real-world events—faults, outages, regulatory changes, staffing constraints—introduce dynamics the model was never calibrated for.
At this point, the digital twin faces a choice:
- Simplify reality to preserve coherence
- Or absorb complexity and lose usability
Most systems choose the former, quietly drifting away from reality.
Why “More Data” Doesn’t Fix the Problem?
A common response to digital twin underperformance is to add:
- More sensors
- Higher-resolution data
- Faster update cycles
This treats the problem as one of insufficient fidelity. In practice, the problem is one of misplaced precision.
More data does not resolve:
- Ambiguous accountability
- Informal decision-making
- Organisational workarounds
- Conflicting objectives
A perfectly instrumented system can still behave unpredictably when human and institutional factors dominate outcomes.
Digital twins struggle not because they lack data, but because they cannot encode intent, judgement, and politics.
The Hidden Cost of False Confidence
Perhaps the most dangerous failure mode of digital twins is not that they are wrong—but that they appear right. A clean dashboard, a responsive simulation, and confident forecasts can create a false sense of control. When reality diverges, the failure is often blamed on execution rather than on the model itself.
In infrastructure, this matters because:
- Decisions affect public safety
- Failures have cascading consequences
- Recovery is slow and visible
A misleading model can be worse than no model at all.
A Timeless Pattern, Not a Temporary One
Digital twins are not the first InfraTech concept to encounter this wall. Similar cycles have played out with:
- Expert systems
- Centralised optimisation platforms
- Smart city command centres
- Fully automated grid control
Each wave promised system-level mastery. Each underestimated the entropy of lived operations.
The lesson is not that digital twins are useless. It is that their usefulness is context-bound.
Where Digital Twins Actually Add Value?
Digital twins tend to succeed when they are:
- Used for design-time analysis, not live control
- Applied to subsystems, not entire networks
- Treated as decision support, not decision authority
- Explicitly scoped to known uncertainties
They fail when positioned as:
- A real-time mirror of complex infrastructure
- A replacement for human judgement
- A control layer over institutional processes
The Deeper Insight
The recurring failure of digital twins in live infrastructure is not a technological shortcoming. It is a category error.
Infrastructure systems are not machines to be perfectly mirrored. They are adaptive, contested, and evolving systems. Which leads to the central insight of this framework:
Digital twins struggle in live infrastructure not because they are too simple, but because reality is too entropic.
Until InfraTech acknowledges this openly, digital twins will remain trapped in a familiar cycle—impressive in theory, constrained in practice, and perpetually one iteration away from relevance.
Sober evaluation begins by accepting that some systems cannot be fully simulated—and must instead be governed, observed, and adapted to, not digitally replicated.
That is not a failure of imagination. It is a mark of infrastructural maturity.