“Digital twin” is one of those terms that can mean everything and nothing. From the 3D model of a turbine to an Excel export with a timestamp, a lot of things are called that today. In this article we sort it out with concrete examples — and with an honest assessment of which of them are real twins and which are “only” digital shadows.

Digital shadow or digital twin — where is the difference?

The yardstick is simple and comes from research: if data flows automatically from the original into the replica, it is a digital shadow — the replica shows what happens in the plant, but never intervenes. If data also flows back automatically — meaning the replica sets setpoints or switching commands on the plant — it is a digital twin.

Digital shadowDigital twin
Data flowPlant → replica, automaticPlant → replica → plant, automatic in both directions
EffectVisibility: state, KPIs, alarmsIntervention: setpoints, schedules, switching commands
Typical state in mid-sized companiesmost projectsenergy twin with a return channel within safety limits
Sensible entry point?yes — replaces gut feeling with measured valuesyes, once the shadow is in place and the benefit of the intervention is measurable

Of the examples below, only the energy twin meets this full definition — and that is fine: a good shadow is the prerequisite for every twin.

The examples at a glance (as of 2026)

ExampleData basisShadow or twin?Typical benefit
Asset twinoperating states, temperatures, vibrationsmostly shadowmaintenance by condition instead of calendar
Process twincycle times, unit counts, scrap (OEE)mostly shadowperformance loss visible the same day
Energy twinload profiles, generators, controllable consumerstwin (return channel within safety limits)shave peak loads — e.g. €45,000 grid fees per year
Product twinprocess parameters, test results per partshadow (record)traceability, digital product passport
Building twinheat, electricity and water meters of all sitesmostly shadowspot outlier sites immediately

Example 1: The asset twin — condition instead of calendar

The most common pattern in practice: machines and units deliver operating states, temperatures, vibrations and operating hours into a living replica. Maintenance plans according to actual condition instead of a fixed calendar — failures announce themselves in the replica before the machine stops.

Honest assessment: in most plants this is initially a digital shadow — data flows in, maintenance is still done by people. That is perfectly fine: the shadow alone replaces gut feeling with measured values. The discipline behind it is condition monitoring: tracking the health of pumps, compressors and drives from their own operating data.

Example 2: The process twin — the line in a live picture

Here the software maps the material flow of a production: cycle times, unit counts, scrap rates, bottlenecks. The leading KPI is overall equipment effectiveness (OEE), measured per line and shift. Improvements are first tested in the replica — which shift runs most efficiently, where does the scrap come from? — and then implemented.

The value lies in reaction time: instead of learning in the monthly report that line 2 has lost output, plant managers and supervisors see it the same day. We observe in projects that this visibility alone changes behavior — before anything is automated. How this works on a platform is described on our page about production management software.

Example 3: The energy twin — the replica that saves money

Our favorite for getting started, because the benefit is directly measurable in euros. The replica covers the energy flows of the plant: load profiles, generators such as PV or CHP units, controllable consumers. Two stages run on top of it:

  • Stage 1 (shadow): see load profiles and energy per unit live, detect deviations, automate reports — the same data basis that an ISO 50001 implementation needs.
  • Stage 2 (twin): the software forecasts the peak load from the live load profile and the production plan, tests switching strategies on the replica — and executes the released strategy automatically. Within safety limits: control only happens inside the boundaries that the people responsible for the plant have defined beforehand.

This makes the energy twin one of the few practical patterns that meets the full twin definition — data flows in both directions. What this means economically, in the example from our active energy management calculation: 300 kW of shaved peak at a demand charge of €150 per kW and year saves €45,000 in grid fees per year. How the forecast works in detail is explained in our article on peak load management with AI control.

Example 4: The product twin — the digital record of every part

Every manufactured part gets a digital counterpart: process parameters during manufacturing, test results, later the maintenance history in the field. For manufacturers with traceability obligations — food, medical technology, automotive — the pattern has long been everyday practice, even if it is rarely called a “digital twin” there.

It becomes exciting through standardization: the asset administration shell (AAS, IEC 63278) defines how such product and asset records are exchanged across manufacturers — from the digital nameplate to the CO₂ footprint.

Example 5: The building twin — properties in the replica

The pattern also works outside manufacturing: anyone operating many buildings — multi-site retailers, municipalities, facility managers — maps the heat, electricity and water meters of all sites in one replica. Outliers stand out immediately: the heating that runs through in summer, the site with twice the base consumption. The data basis is usually delivered by existing M-Bus and Modbus meters — new metering equipment is rarely needed.

The common thread: data first, then the twin

All examples share the same core — and the same most frequent wrong decision. The core: a twin is only as good as its data connection. The wrong decision: starting with the model instead of the data.

For plants with existing equipment, the economical path therefore reads:

  1. Connect existing sources — PLCs via OPC UA, meters via Modbus, sensors via MQTT. Without intervening in the controllers.
  2. Structure the replica — plants, areas, KPIs. That is the actual twin, long before anything appears in 3D.
  3. Open the return channel where it pays off — first with non-critical, controllable consumers, within safety limits and released by you.

This is exactly the pattern we use to build twins with eco2lot — the industrial IoT platform connects the existing sources, and the eco2lot areas Energy and Production provide the structured replica on top.

Frequently asked questions

What is the difference between a digital shadow and a digital twin?
With a digital shadow, data flows automatically from the plant into the replica — it shows the state, but never intervenes. With a digital twin, data also flows back automatically: the replica acts on the plant, for example by setting setpoints or switching commands. Most industrial projects today are shadows; the energy twin with a return channel within safety limits is the most common example of a real twin.
What is the difference between a digital twin and a dashboard?
A dashboard displays data — it is the presentation layer. A digital twin is the structured replica of the plant behind it, continuously updated with live data and, in its full form, acting back on the plant. Every twin project has dashboards; but not every dashboard has a twin behind it.
Which example is the best starting point for a mid-sized plant?
In our projects usually the energy twin: the data basis (meters, load profiles) is almost always available, the benefit is directly measurable in euros — for example through avoided peak loads — and the return channel can be limited, within safety limits, to non-critical consumers. The production twin with OEE often follows as the second step on the same platform.
Does a digital twin need a 3D model?
For most operational applications: no. What matters is the data model — which plant delivers which measured values, how they relate to each other, which KPIs result from them. A 3D model makes the replica more vivid, but changes nothing about its significance. Anyone who starts with 3D instead of data often builds an expensive shell without content.
How long does it take to build a digital twin?
For a first, usable twin of one plant section we calculate 4 to 8 weeks: connect data points (OPC UA, Modbus, MQTT), define structure and KPIs, set up alarms. The expansion — more lines, more sites, the automatic return channel — follows step by step afterwards, wherever a measurable benefit shows.

Back to the blog