Pragmatic Digital Twins: Start With One Purpose, Not a Replica of Everything

Executive summary

"Digital twin" has become one of the most elastic terms in construction. An analysis of more than 15,000 scientific publications found that definitions shift by domain and scale, and that the features most often promised — simulation, AI, real-time data, two-way control — are not yet mature in built-environment twins [2]. The international vocabulary standard, ISO/IEC 30173:2023, is deliberately modest: a digital representation of a target entity, connected by data so that the physical and digital states converge "at an appropriate rate of synchronization" [1]. Nothing in that definition requires a whole-building replica or real-time everything. The trend worth following is the opposite of the hype: twins scoped to one purpose, one system, one decision — most often HVAC and energy — where the evidence of value is strongest and the data foundations are achievable.

The real problem and what's changing

The first generation of digital-twin pitches sold a living replica of the whole asset: every system modelled, every sensor streaming, the model always true. Few of those exist. What a 2026 systematic review found in the literature of operational twins is narrower and more useful: of 46 deployments that met inclusion criteria (from 2,839 records), 34 focused on indoor environmental quality and energy, largely HVAC and building automation [6]. Only about half documented formal semantics for the building, and fewer than a third used them to actually shape analysis or decisions. Outputs mostly stayed inside each platform's own interface, with limited evidence of integration into maintenance systems or control sequences [6].

That is the real state of the art: capable point solutions that do not yet transfer across sites. The shift underway is to stop treating "the twin" as a product you buy and start treating it as a question you answer — which decision, for which system, on which data, updated how often.

Why now

Three things converge. First, the vocabulary has settled: ISO/IEC 30173:2023 gives a sector-neutral definition that separates the representation, the data connection and the synchronization rate [1] — which makes it legitimate to build a twin that updates monthly if that is what the decision needs. Second, the maturity conversation has matured. The Atkins/IET spectrum of six elements (0 to 5) — from reality capture, through object models with metadata, to prediction and decision support, two-way exchange and finally autonomous operation — gives teams a vocabulary for choosing a level instead of defaulting to the top [4]. Third, the evidence base is finally telling us where twins pay off: the HVAC-and-energy concentration in deployed systems [6] matches where the measurable results are.

The business pressure is also familiar from the rest of this series: McKinsey describes the probable evolution of BIM as a continuously updated view of real-world state that supports approvals and payments with documented evidence [12]. A twin is the operational form of that idea — but only if its data can be trusted.

How it works and where it applies

A pragmatic twin has a stable shape: purpose (one decision or KPI — comfort complaints, peak demand, equipment faults) → scope (the systems and spaces that decision depends on, not the whole asset) → representation (the model, at the detail the purpose needs — often a coordinated mechanical model, not an architectural visualisation) → data connection (which signals, from which BAS or sensors, at which rate) → analysis (monitoring, fault detection, optimisation, or flexibility provision — the four service areas a 2026 HVAC+R review identifies [9]) → action (a work order, a setpoint, a report — delivered into the system people already use).

Concrete applications, in order of evidence:

• HVAC monitoring and fault detection: the most common deployed use [6], and the area the HVAC+R review treats as closest to practice [9].
• Energy optimisation of central plant: a Hong Kong hotel study linked a chiller-plant twin fed by 1,416 IoT sensors (17,520 hourly records) to a staged controller; the AI-driven mode used about 21.9% less daily HVAC energy than fixed operation, though the authors did not assess payback [10].
• Handover-to-operations twins: a coordinated mechanical model plus identified assets, feeding the owner's maintenance system — the use case where ISO 19650-3's operational information requirements apply directly [11].
• Flexibility and demand response: emerging; the review lists it as a service area but with thinner real-building evidence [9].

What is not on the list is telling: whole-building, all-discipline, real-time replicas. The literature's two recurring twin types are "high-performance real-time" and "long-term decision support" [2] — and most building decisions belong to the second.

Business and team impact

The economic logic of a focused twin is the same as the data-architecture argument earlier in this series: the cost of initial definition (identity, semantics, data mapping) is high, the cost of each additional use is low. Scoping narrowly keeps the first cost survivable. It also makes the twin measurable: a twin with one KPI has a baseline, so it can be shown to work or to fail — which most "platform" twins never are.

For BIM and VDC teams, the implication is concrete. The model that becomes the twin's backbone is usually the coordinated mechanical model, and its usefulness downstream depends on things decided during design and construction: stable asset identifiers, properties referenced to a dictionary, equipment service clearances modelled as geometry, and a handover package structured for the operator rather than for the contract. The 2026 review names machine-readable building context and lifecycle-governed metadata as the limiting factors for portability and for structured access by AI agents [6]. That is modelling work, and it is done — or not — long before anyone buys a twin platform.

The organisational finding is blunter. A 2026 synthesis of 47 studies on healthcare facility management concludes that stalled adoption is "a structurally produced outcome" of separating capital delivery from operations, unclear data ownership, and excluding facility managers from early stages — constraints the authors judge to outweigh technical immaturity [7]. Twins fail at the handover seam, not in the software.

Barriers, risks, and maturity level

Definition drift is the first risk: if a team cannot say which element of the maturity spectrum it is targeting [4], it will buy features it will not use. Modularity and users are the second: a 2025 review in Construction Management and Economics flags technological complexity, lack of modularity and low user interaction during development, and notes that stakeholder perspectives are far less studied than technical work [8]. Data foundations are the third: the HVAC+R review locates the "modelling bottleneck" at the data level (harmonisation, ontologies), the process level (behaviour prediction and service integration) and the integration level (no standard twin architectures) [9]. Survivorship bias is the quiet fourth: the healthcare synthesis warns that failed or discontinued twin projects are under-reported, so published maturity is probably overstated [7].

Maturity, honestly stated: monitoring and fault-detection twins for HVAC are established in research and increasingly in practice; prediction and optimisation twins are growing, with measured single-site results [10] but little cross-site transfer [6]; two-way control and autonomous operation are emergent at best, and the 15,000-paper analysis finds bidirectional data flow not yet mature in the built environment [2]. A useful self-check comes from the Gemini-based maturity model, which grades twins across purpose, trust and function on 27 rubrics and places most real projects well below its top "instructive and intelligent" level [5].

How to prepare

1. Write the purpose in one sentence, with a KPI and a baseline — the Gemini Principles put purpose (public good, value creation, insight) before trust and function for a reason [3]. If you cannot name the decision, you are not ready for a twin.
2. Pick a maturity element deliberately [4] and say which rate of synchronisation the decision actually needs [1]; "real-time" is a cost, not a virtue.
3. Scope to one system family — for most buildings, mechanical: that is where deployed twins cluster [6] and where measured results exist [10].
4. Fix the data foundations in the BIM, not in the platform: stable asset identifiers, dictionary-referenced properties, service clearances as geometry, and a handover structured to ISO 19650-3 [11] — see the earlier articles on data architecture and openBIM.
5. Decide the integration target before buying anything: which maintenance system or control sequence will receive the twin's output. Twins whose output stays in their own dashboard are the ones the literature finds do not scale [6].
6. Budget for the twin's maintenance and name an owner; a twin whose data drifts from the building is worse than no twin, because it is believed.

Future outlook and conclusion

The likely direction is federation rather than consolidation: several narrow twins — plant, ventilation, water — sharing identifiers and semantics, with AI assistants reading structured building context rather than replacing it [6][12]. The autonomous, self-managing building at the top of the maturity spectrum remains speculative; the evidence does not yet support it, and the organisations that reach the middle levels will be the ones that scoped tightly first.

The strategic point for owners and for the BIM teams that serve them: a digital twin is not the model made live; it is a decision made measurable. Start with the decision. The model — if it was built with identity, semantics and handover in mind — will be ready for it.

Frequently asked questions

Is a BIM model a digital twin? No. In the maturity spectrum a coordinated model with metadata is element 2; a twin needs a data connection that keeps physical and digital states converging [1][4]. The model is the backbone a twin is built on — which is why its data quality matters.

Does a digital twin have to be real-time? Only if the decision it serves is. ISO/IEC 30173 speaks of an appropriate synchronisation rate [1], and the literature distinguishes real-time twins from long-term decision-support twins, which most building questions fall under [2].

Where should a mechanical contractor or engineering firm start? With the system you already model best and the data you already have: HVAC monitoring and fault detection is where deployed twins concentrate and where results are measured [6][9][10]. Make sure the coordinated model carries stable identifiers and dictionary-referenced properties so it can become that twin's foundation.

---

Sources and references

[1] ISO/IEC. ISO/IEC 30173:2023 — Digital twin — Concepts and terminology. ISO/IEC JTC 1/SC 41, published 7 November 2023.

[2] Abdelrahman, M.; Macatulad, E.; Lei, B.; Quintana, M.; Miller, C.; Biljecki, F. "What is a Digital Twin Anyway? Deriving the Definition for the Built Environment from over 15,000 Scientific Publications." Building and Environment, vol. 274, 112748, 2025 (preprint arXiv:2409.19005).

[3] Centre for Digital Built Britain (Bolton, A.; Enzer, M.; Schooling, J. et al.). "The Gemini Principles." University of Cambridge, 2018.

[4] Atkins / The Institution of Engineering and Technology (Evans, S. et al.). "Digital twins for the built environment: An introduction to the opportunities, benefits, challenges and risks." 2019 — digital twin maturity spectrum, elements 0–5.

[5] Chen, L.; Xie, X.; Lu, Q.; Parlikad, A. K.; Pitt, M.; Yang, J. "Gemini Principles-Based Digital Twin Maturity Model for Asset Management." Sustainability, 13(15), 8224, 2021. https://doi.org/10.3390/su13158224

[6] Elias, C.; Hutchins, I.; Issa, R. R. A. "Operational digital twins in the built environment: A systematic review of semantic infrastructure and deployment practice." Journal of Information Technology in Construction (ITcon), vol. 31, pp. 1015–1041, 2026. https://doi.org/10.36680/j.itcon.2026.043

[7] Almukhtar, A.; Shahzad, M.; Tah, J. H. M. "From BIM to digital twins in public healthcare facility management: bridging the gap between technological potential and operational reality." Frontiers in Built Environment, vol. 12, 2026. https://doi.org/10.3389/fbuil.2026.1828585

[8] Miraj, P.; Wang, T.; Koutamanis, A.; Chan, P. "Organising digital twin in the built environment: a systematic review and research directions on the missing links of use and user perspectives of digital twin in the AEC sector." Construction Management and Economics, 43(6), pp. 465–481, 2025. https://doi.org/10.1080/01446193.2025.2451631

[9] Ibrahim, N. I.; Vering, C.; Giannetti, N.; Kim, M.; Sholahudin; Storek, T.; Müller, D.; Saito, K. "State-of-the-art of the digital twin concept in HVAC+R systems for thermal management of buildings." Renewable and Sustainable Energy Reviews, vol. 229, 116601, 2026. https://doi.org/10.1016/j.rser.2025.116601

[10] Wong, R. W. M.; Loo, B. P. Y. "Improving Energy Efficiency in Hotels Through Evolutionary Digital Twin Technology: A Conceptual Framework and Empirical Evidence." Buildings, 15(24), 4406, 2025. https://doi.org/10.3390/buildings15244406

[11] ISO. ISO 19650-3:2020 — Information management using building information modelling — Part 3: Operational phase of the assets.

[12] Ahmoye, D.; Sjödin, E.; Blanco, J. L.; Mawji, A.; Winslade, M.; Rogers, P. "How AI is reshaping the future of the AEC industry." McKinsey & Company, July 15, 2026.

Traceability note: no market-size figures are used — commercial estimates for "digital twins in construction" differ by more than a factor of two depending on scope. Vendor claims of 10–40% energy savings were excluded for lack of a primary source. A fault-detection accuracy figure (">95%") attributed to the HVAC+R review in a secondary summary could not be confirmed in the paper's text and was left out. The hotel result [10] is a single site; the 21.9% figure is taken from the authors' stated limitations and a secondary review, as the full results section was not accessible.