Digital Twin in Manufacturing: Current Capabilities and Applicability Limits
Critical analysis of digital twin maturity levels, practical applications, economic viability, and realistic limits in modern manufacturing.
Serhiy Andrejev
Lead Machine Learning Developer
The digital twin concept has attracted extraordinary investment and attention across the manufacturing sector. Yet in practice, the gap between vendor promises and operational reality remains substantial. This article offers a critical, experience-based assessment of what digital twins can deliver today, where they fall short, and how manufacturers should approach adoption without overspending on unrealized potential.
Introduction: The Most Overrated Concept of Industry 4.0
Few terms in the industrial technology lexicon carry as much marketing weight — and as little operational clarity — as "digital twin." The global digital twin market reached approximately $24.5 billion in 2025 and is projected to exceed $328 billion by 2033, growing at a compound annual rate above 35% [3]. Technology vendors routinely present digital twins as the centerpiece of smart manufacturing: a complete, real-time virtual replica of a factory that autonomously optimizes itself.
The reality on the ground is far more nuanced. In our experience working with manufacturing enterprises across Ukraine and the EU, most organizations that claim to have a "digital twin" actually operate a 3D visualization of their facility connected to a handful of sensors — a useful tool, but fundamentally different from what the term implies in its full technical sense. According to McKinsey's survey of senior industrial executives, while 86% stated that digital twins were applicable to their organizations, fewer than half had implemented even a basic version [2].
This disconnect matters because it leads to misallocated budgets, inflated expectations, and — most critically — delayed adoption of narrower applications that deliver proven value. The purpose of this article is to separate substance from hype: to define what a digital twin actually is, classify its maturity levels, identify where it works today, acknowledge where it does not, and provide a realistic economic framework for adoption decisions.
What Is Actually Behind the Term
The ISO 23247:2021 standard series provides the most authoritative framework for understanding digital twins in manufacturing. Published across four parts, this standard defines a digital twin as a fit-for-purpose digital representation of an observable manufacturing element with synchronization between the element and its digital representation [1]. The framework covers overview and general principles (Part 1), reference architecture (Part 2), digital representation of manufacturing elements (Part 3), and information exchange protocols (Part 4) [6].
A critical distinction — often overlooked in commercial literature — separates three related but fundamentally different concepts. A digital model is a digital representation with no automated data exchange between the physical object and the model. A digital shadow features one-directional automated data flow from the physical object to the digital representation. A digital twin, in the strict sense, requires bidirectional automated data flow: the physical object informs the digital representation, and the digital representation can trigger changes in the physical object [6].
In our assessment, this distinction is not academic. It determines the actual capability of the system and, consequently, the investment required to build and maintain it. Most implementations marketed as "digital twins" are, by this definition, digital shadows at best — they consume sensor data and display it, but they do not close the loop by automatically influencing the physical process. Recognizing where an implementation sits on this spectrum is the first step toward realistic expectation-setting.
Four Maturity Levels of Digital Twins
Based on our analysis of the ISO framework, IBM's digital twin maturity model [5], and practical implementation patterns observed across manufacturing sectors, we propose a four-level maturity classification that reflects both technological capability and organizational readiness.
Level 1: Static Model
The entry point. A static digital twin is essentially a detailed 3D CAD or BIM model enriched with engineering documentation and as-built specifications. There is no live data connection. The model serves design validation, spatial planning, and documentation purposes. Most manufacturing enterprises that report "having a digital twin" operate at this level.
Level 2: Connected Model
At this level, the digital representation receives real-time data from IoT sensors, SCADA systems, or MES platforms. The connection is unidirectional — physical-to-digital — making this a digital shadow rather than a true twin. However, the practical value is significant: operators gain real-time visibility into equipment status, environmental conditions, and process parameters. Anomaly detection becomes possible through threshold monitoring and statistical analysis.
Level 3: Dynamic Simulation
The model can now run simulations using both historical data and real-time inputs. Physics-based engines or discrete event simulation tools allow operators to test "what-if" scenarios: What happens if we increase throughput on Line 3 by 15%? What is the thermal impact of running Furnace B at maximum capacity for 72 hours? This level requires substantially more computational infrastructure — typically cloud or high-performance computing resources — and domain expertise in simulation modeling.
Level 4: AI-Augmented Twin
The most advanced level integrates machine learning and AI capabilities into the simulation layer. The twin not only reflects current state and simulates scenarios but also predicts future states and prescribes optimal actions. At this level, bidirectional data flow becomes essential: the twin's recommendations must be actionable and, in some cases, automatically executed. This is the only level that meets the strict ISO definition of a digital twin.
Table 1. Digital Twin Maturity Levels: Requirements, Capabilities, and Adoption
| Dimension | L1: Static Model | L2: Connected Model | L3: Dynamic Simulation | L4: AI-Augmented Twin |
|---|---|---|---|---|
| Data requirements | CAD/BIM, as-built docs | Real-time sensor data (IoT) | Historical + real-time streams | All prior + ML training datasets |
| Infrastructure | File storage, basic 3D viewer | SCADA/MES integration, cloud/edge | HPC or cloud compute, physics engines | AI/ML platform, data lake, edge AI |
| Key capabilities | Design validation, documentation | Monitoring, anomaly detection | What-if scenarios, process optimization | Predictive/prescriptive analytics, autonomous adjustments |
| Typical cost range | $50K–$200K | $200K–$1M | $1M–$5M | $5M–$20M+ |
| Implementation timeline | 1–3 months | 3–9 months | 9–18 months | 18–36 months |
| Candidate industries | All manufacturing | Process, discrete with critical assets | Automotive, aerospace, pharma | Semiconductor, energy, large-scale process |
| Estimated adoption rate | ~60% of large manufacturers | ~35% | ~10–12% | <3% |
| ISO classification | Digital model | Digital shadow | Advanced digital shadow | Digital twin (bidirectional) |
Sources: IBM Digital Twin Maturity Model [5]; McKinsey manufacturing survey [2]; Grand View Research market data [3]; author's estimates based on implementation experience.
Figure 1. Digital twin adoption rate vs. demonstrated cost reduction by maturity level. Higher maturity yields greater impact but far lower adoption. Data from McKinsey [2], Grand View Research [3], and author's estimates.
Where It Works Today
Despite the legitimate criticism of digital twin hype, several application domains have demonstrated consistent, measurable value. Based on published research and our own project observations, three areas stand out.
Product design and virtual prototyping. In automotive and aerospace manufacturing, digital twins of individual components and assemblies have reduced physical prototyping cycles by 20–50%. This is the most mature application, closest to the original NASA concept of maintaining a digital replica of spacecraft systems. The economics are compelling because physical prototypes are expensive, and iteration speed directly affects time-to-market [2].
Critical equipment monitoring. For high-value, failure-sensitive assets — gas turbines, large compressors, CNC machining centers, industrial boilers — connected digital twins (Level 2–3) deliver proven ROI. McKinsey reports that predictive maintenance enabled by digital twins reduces unplanned downtime by 30–50% and maintenance costs by 40–60% [2]. In our practice, we have observed similar results for enterprises operating continuous processes where a single hour of unplanned downtime costs tens of thousands of euros.
Personnel training and simulation. An increasingly validated use case involves creating digital replicas of production environments for operator training. This application is particularly valuable in hazardous industries (chemical processing, metallurgy, energy) where hands-on training carries inherent risk. The digital twin here functions primarily at Level 3 — a simulation environment that does not require bidirectional control but does require accurate physics and process modeling.
Deloitte's 2025 Smart Manufacturing Survey confirms these patterns: companies implementing smart manufacturing technologies, including targeted digital twins, report 10–20% improvement in production output and 7–20% improvement in employee productivity [4].
Where It Does Not Work Yet
Equally important is an honest assessment of the boundaries. Several widely promoted applications remain aspirational rather than operational.
The "entire factory" digital twin. The vision of a comprehensive, real-time digital replica of an entire manufacturing facility — from raw material intake to finished goods — remains largely unrealized. The fundamental challenge is data integration: a typical mid-size factory operates dozens of heterogeneous systems (SCADA, MES, ERP, quality management, warehouse management) with incompatible data models, communication protocols, and update frequencies. Fraunhofer IPK's study of 26 manufacturing companies in the DACH region confirmed that cross-system and cross-company digital twin networking has not yet been achieved at meaningful scale [5].
Autonomous optimization. The promise that a digital twin will autonomously adjust production parameters without human intervention remains confined to narrow, well-bounded processes (such as temperature control in a single furnace). Extending autonomous control to multi-stage production lines introduces complexity that current AI systems cannot reliably manage. The risk-reward calculus simply does not favor removing human oversight from critical manufacturing decisions — not yet.
Scaling beyond pilots. Perhaps the most telling statistic: approximately 75% of organizations implementing digital twins report difficulty scaling beyond pilot projects [2]. The pilot works — monitoring three machines on one line — but extending it to 300 machines across 12 lines encounters exponential growth in data management complexity, integration challenges, and maintenance overhead. This scaling gap is the single largest barrier to digital twin adoption in manufacturing today.
Economics: Cost of Creation vs. Expected Impact
The economic case for digital twins is highly dependent on the maturity level targeted and the specificity of the application. Broad generalizations about ROI are misleading; the variance between a well-scoped Level 2 implementation and an overly ambitious Level 4 project can be an order of magnitude in both cost and risk.
Table 2. Digital Twin Economics by Maturity Level
| Economic Factor | L1: Static Model | L2: Connected Model | L3: Dynamic Simulation | L4: AI-Augmented Twin |
|---|---|---|---|---|
| Initial investment | $50K–$200K | $200K–$1M | $1M–$5M | $5M–$20M+ |
| Annual maintenance | 5–10% of initial | 15–20% of initial | 20–25% of initial | 25–35% of initial |
| Typical ROI timeline | Indirect (design efficiency) | 12–18 months | 18–36 months | 36–60 months |
| Demonstrated cost reduction | N/A | 10–15% maintenance savings | 15–25% process optimization | 20–30% operational costs |
| Risk of failure | Low | Low–moderate | Moderate–high | High |
| Data team required | No | 1–2 engineers | 3–5 specialists | 8–15+ (data, ML, domain) |
Sources: McKinsey [2]; NIST economic analysis; Grand View Research [3]; author's project estimates.
NIST estimates that if digital twins were fully adopted across US manufacturing, they could generate $37.9 billion in annual value [6]. However, "fully adopted" is a theoretical ceiling — not a near-term expectation. The more actionable insight is that targeted implementations at Level 2–3, focused on high-value assets or critical process bottlenecks, consistently deliver positive ROI within 18 months.
In our assessment, the most common economic mistake is attempting to build a Level 4 system from scratch. The phased approach — start with a Level 2 connected model for the highest-value asset, demonstrate measurable impact, then incrementally expand scope and capability — consistently outperforms the "boil the ocean" strategy both in financial returns and organizational adoption.
Author's Assessment and Conclusions
The digital twin is not a binary proposition — it is a spectrum. Understanding where an organization sits on this spectrum, and where it realistically needs to be, is the foundation of any sound adoption strategy.
Several conclusions emerge from our analysis and practical experience:
Most manufacturers are at Level 1–2, and that is acceptable. The pressure to achieve a Level 4 AI-augmented twin is driven more by vendor marketing than by operational necessity. A well-implemented Level 2 connected model — providing real-time monitoring and anomaly detection for critical assets — delivers substantial, measurable value for a fraction of the Level 4 cost.
The maturity gap is widening, not narrowing. Despite market growth, the gap between what vendors promise and what most manufacturers can absorb is increasing. The limiting factor is rarely technology — it is data readiness, organizational capability, and integration maturity. Enterprises that invest in data infrastructure and cross-system integration will be far better positioned when Level 3–4 capabilities become more accessible.
The convergence of AI and IoT will accelerate progress, but not uniformly. Advances in edge computing, foundation models for time-series data, and low-code ML platforms are reducing the technical barrier to Level 3–4 implementations. However, the organizational and data-quality barriers remain. We expect meaningful Level 3 adoption (above 25% of large manufacturers) within three to five years, while Level 4 will remain limited to specific high-value processes in capital-intensive industries.
The recommended approach is phased and evidence-driven. Begin with the asset or process where downtime costs are highest. Implement a Level 2 connected model. Measure results over 6–12 months. Use demonstrated ROI to justify expansion — both in scope (more assets) and in depth (advancing toward Level 3). This is not a conservative strategy; it is the strategy with the highest probability of delivering actual value rather than an impressive demonstration that never scales.
The digital twin, stripped of its marketing veneer, is a powerful engineering tool with well-defined capabilities at each maturity level. The challenge for manufacturing leaders is not whether to adopt it, but how to adopt it at the right level, for the right application, with realistic expectations and disciplined economics.
References
- ISO. "ISO 23247-1:2021 — Automation Systems and Integration — Digital Twin Framework for Manufacturing — Part 1: Overview and General Principles." 2021. https://www.iso.org/standard/75066.html
- McKinsey & Company. "Digital Twins: The Next Frontier of Factory Optimization." 2024. https://www.mckinsey.com/capabilities/operations/our-insights/digital-twins-the-next-frontier-of-factory-optimization
- Grand View Research. "Digital Twin Market Size and Share Report, 2033." 2025. https://grandviewresearch.com/industry-analysis/digital-twin-market
- Deloitte. "Industry 4.0 and the Digital Twin Technology." 2024. https://www.deloitte.com/us/en/insights/industry/manufacturing-industrial-products/industry-4-0/digital-twin-technology-smart-factory.html
- IBM. "Digital Twin Maturity Model." 2024. https://www.ibm.com/think/topics/digital-twin-maturity-model
- NIST. "An Analysis of the New ISO 23247 Series of Standards on Digital Twin Framework for Manufacturing." 2022. https://www.nist.gov/publications/analysis-new-iso-23247-series-standards-digital-twin-framework-manufacturing
- ARC Advisory Group. "The Digital Twin in Industry and Infrastructure." 2024. https://www.arcweb.com/industry-best-practices/digital-twin-industry-infrastructure