Building a Digital Twin system — from business problem to operation & scale-out
This guide covers all 8 implementation stages, the decision points (stage gates), a reference architecture, how the project is organized, and a time and resource estimator you can adjust to your own scale.
What is a Digital Twin, and at what level should you start?
A Digital Twin is a digital replica kept up to date with real data to reflect the state and/or behavior of a system. A 3D model alone is not a Digital Twin; what matters is connected data and a model with business meaning. Choose the level by business value and data quality, not by technology.
5 implementation principles
8 end-to-end implementation stages
Stages can overlap, but each one has clear deliverables and a gate. Budget expands only after value and data access have been proven. Select a stage to see details; the time and effort on the right are calculated from the configuration in the Estimate section.
Stage gates: GO / HOLD / STOP decisions
Each gate signs off on deliverables to avoid the situation where “the code is done but can't be accepted”.
| Gate | After stage | Pass criteria | Signed off by |
|---|
Agreed thresholds met → move to the next stage and release the matching payment milestone.
Data cleanup, additional tags, access rights, or clearer requirements are needed; a remediation plan and review date are set.
Not feasible with current data or constraints; stop early to protect the budget and record lessons learned.
6-layer reference architecture
A vendor-neutral architecture that can run in the cloud, on-premise, or hybrid depending on the OT/IT environment. Select a layer to see its function, typical technologies, and design notes.
Estimated time & resources
Pick a scale to load a sample configuration, then fine-tune it to your situation. Time and effort are driven by scope of work (number of assets, protocols, integrations, solution level), not by factory floor area.
Implementation roadmap (Gantt)
Hover for details · click to open a stageEffort by stage
man-monthsAcceptance milestones by gate
ProposedQuick comparison by scale (sample configurations)
Typical durations: PoC 4–8 weeks · Pilot 2–4 months (after design) · Single site 4–9 months · Enterprise 6–12+ months. Excludes waiting time for procurement, approvals, or lengthy OT access provisioning.
How this compares with leading vendors' methods and figures
The process and timelines in this guide are checked against public material from vendors (Siemens, GE Vernova, AVEVA, Bentley, Dassault, AWS, Microsoft, NVIDIA), consultancies (McKinsey, Deloitte, Capgemini, IDC), standards (ISO 23247, Digital Twin Consortium, NIST), and Vietnamese market data.
Implementation methods
| Source | Published steps / framework | Mapping to the 8-stage process |
|---|---|---|
| McKinsey | Blueprint → Base twin (built in 3–6 months, starting from a core data product and 1–2 use cases) → Boost capabilities | Design → Prove + Build → Operate & scale |
| Deloitte | Imagine → Identify (high-value, high-likelihood pilot) → Pilot → Industrialize → Scale → Monitor & measure | Qualify/Discover → Assess/Design → Prove → Build/Deploy → Operate |
| GE Vernova | Define purpose → form an IT–operations–user team → assess technology readiness → integrate data, digital thread → build, validate, test the model → add ML with domain experts → ensure interoperability → deploy | Covers Qualify, Discover, Assess, Build, Prove, Deploy; Design and Operate are not separate steps |
| AVEVA | Lifecycle Design → Build → Operate → Optimize | Design → Build/Deploy → Operate |
| Digital Twin Consortium | Capabilities Periodic Table: ~60 capabilities in 6 groups (Data services, Integration, Intelligence, UX, Management, Trustworthiness), used to define requirements per use case | A tool for Discover → Assess → Design |
| ISO 23247 | 4-domain reference architecture: observable manufacturing elements · device communication · digital twin · user | Matches the Physical/OT → Connectivity → Twin model → Experience layers |
| AWS | Maturity levels L1 Descriptive → L2 Informative → L3 Predictive → L4 Living | Roughly equivalent to the 4 levels Visual → Operational → Predictive → Prescriptive |
| NIST IR 8356 | Security and trust issues specific to digital twins (2025) | Secure-by-design principle, gates G2 and G4 |
Published timelines
This guide's estimation model gives PoC ~10 weeks (including discovery), Pilot ~5 months, single plant ~7 months, multi-plant ~13 months — within the ranges above.
Published results
| Source | Result | Type of figure |
|---|---|---|
| IDC (IBM-sponsored) · IBM Maximo | 5-year ROI 450%, 13-month payback, unplanned downtime −43% | Sponsored study |
| AVEVA · Cosmo Oil, SCG Chemicals | Payback < 1 year; maintenance cost −40% | Vendor-reported |
| GE Vernova · SmartSignal | Average ROI reached in 3.41 months | Vendor-reported |
| Bentley · Baosteel | Upgrade schedule 12 → 6 months; rework rate 15% → 0.8% | Vendor-reported |
| Siemens · Calvatis | Downtime −10%, cleaning chemicals −6% | Vendor-reported |
| Capgemini Research Institute (>1,000 organizations) | ~15% improvement in sales and operational metrics; system performance +25% | Independent survey |
| McKinsey (2024) | 5–7% monthly cost savings with an assembly-plant scheduling twin; 44% of companies have deployed a twin | Independent consultancy |
| Siemens · True Cost of Downtime 2024 | Unplanned downtime equals ~11% of revenue for the 500 largest companies | Industry study |
Vendor-reported figures are usually best cases.
Deployments in Vietnam
360° twin of the Urea, Ammonia, NPK, CO₂, Utility units and warehouse; asset tags linked to documents and real-time IoT data. Completed in ~2 months.
ISA-95-based architecture, predictive maintenance for the CDU and RFCC; closed-loop twin pilot at the RFCC planned for June 2027 on Microsoft Fabric.
3D models of telecom stations from drone imagery, AI detection of rust, cracks, and load assessment; processes hundreds of stations per day. Best Digital Transformation Project, Glotel Awards 2025.
Digital twin for urban management covering planning, traffic, construction investment, and flooding; phase 1 from Sep 2026 to Dec 2027, phase 2 through 2030.
Platform and legal notes
- AWS stopped accepting new customers for IoT SiteWise Monitor and the SiteWise Edge Data Processing Pack on 7 Nov 2025; IoT Events reached end of support on 20 May 2026. IoT TwinMaker remains available.
- Azure Digital Twins remains generally available (updated Aug 2026); Microsoft is developing a digital twin builder on Fabric (preview).
- NVIDIA Omniverse is reported to be free for development and production since May 2026. Confirm with NVIDIA.
Recommendation: a vendor-neutral architecture, with the twin model and data layers under the customer's control, avoiding dependence on services that are being scaled back.
- Data Law 60/2024/QH15 (effective 1 Jul 2025) and Decree 165/2025: transferring “important” or “core” data abroad requires a prior impact assessment.
- Cybersecurity Law 116/2025/QH15 takes effect on 1 Jul 2026.
- Personal Data Protection Law 91/2025/QH15 (effective 1 Jan 2026), Decree 356/2025.
- Support: Decree 80/2021 lets SMEs receive up to 50% of digital transformation consulting costs (capped); Decision 2708/QĐ-BCT on smart manufacturing through 2035.
Recommendation: operational data of energy and utility companies should be stored in Vietnam (on-premise or a domestic cloud). This is technical guidance, not legal advice.
Roles & RACI matrix
A Digital Twin needs business, OT, IT, and software expertise. R = Responsible · A = Accountable · C = Consulted · I = Informed. The matrix is finalized at kick-off.
Sponsor — investment decisions, KPIs, priorities, scope.
Process owner / SME — As-Is/To-Be processes, KPI formulas, acceptance.
OT / IT / Data owner — equipment, data, access rights, network, environments.
PM / BA / Architect — planning, requirements, design, delivery coordination.
Data / Backend / 3D / QA — adapters, models, applications, testing, deployment, support.
PLC, sensor, and robot vendors; domain physics/simulation experts; 3D laser scanning providers.
| Activity | Sponsor | Process SME | OT / IT | Nobisoft | Operations |
|---|
* Nobisoft is R for the operations SLA only when Nobisoft provides maintenance under contract; otherwise R belongs to the customer's operations team.
Key risk register
Each risk has an owner, warning signs, mitigation, and a contingency plan from Discovery onward. Select a point on the matrix or in the list.
of companies get key digital use cases into wide-scale deployment; 61% cite unclear ROI as a major barrier
of companies (India survey) are still stuck at PoC/pilot; over 50% see ROI limited by fragmented data
false-positive rate once wiped out all the savings of a predictive maintenance program, so AI models must be validated
Engagement models
Flexible according to the customer's readiness and in-house capabilities.
Discovery / Consulting
Workshops, data assessment, solution blueprint, roadmap, and budget estimate.
Duration: 2–4 weeks
Fixed scope
Fixed price for a defined scope, once acceptance criteria and data access are settled.
Time & Materials
When the backlog is still changing, learning from the pilot, and developing in sprints.
Dedicated team
A team co-developing the platform, connectors, analytics, and multi-plant rollout.
Preparation checklist for the first workshop
Existing documents are enough; you don't need to complete everything before starting. Progress is saved in this browser.