Why the terms get mixed up

'Digital twin' is the term that's stuck, even where something quite different is meant. In German it's often used alongside the English 'digital twin', and in many companies both terms circulate side by side — occasionally for two different things. The term sounds like the future, it appears in funding applications, and suppliers happily apply it to anything that looks three-dimensional.

The result: five people sit in a meeting, say the same word, and mean three different things. One of them is budgeting for two weeks; another for a year.

What matters most for telling these terms apart is the relationship to the real object.

The 3D model: the intended state

A 3D model is designed. It describes how an object is meant to look and be built. It's typically created in engineering as a CAD dataset, long before the product is manufactured. That's where its greatest communication advantage lies: you can show a product that doesn't exist yet.

CAD data isn't a presentation model

This is the most common misunderstanding in this field. Design data is made for manufacturing, not for display. It contains millions of surfaces, standard parts, tolerances, design history and complete internal geometry. Full engineering data like this generally isn't built to run smoothly in a browser.

For sales and marketing, the dataset is therefore prepared, and that's more than a technical conversion. It's reduced to the elements needed to make the point. It's broken down into named assemblies, so things can be opened up and closed again. It gets materials, finishes and motion.

And it's often cleaned up. A design dataset regularly contains components that, for good reason, shouldn't go public: bought-in parts carrying supplier identification, patent-relevant solutions buried inside, manufacturing details a competitor could use to work out how you solved a problem. During preparation, these parts are removed or simplified. What ends up in the finished application is considerably less than what went in. That's technically necessary, and it protects your know-how.

What you need a 3D model for: web product presentation, digital catalogues, configurators, AR views, trade show applications, training content, renderings for marketing material. For practically anything to do with explaining and selling.

The 3D scan: the actual state

A 3D scan is a survey. It captures a real, existing object, using laser scanning or photogrammetry, and initially produces a point cloud from which a surface model is calculated.

The crucial difference from a designed model: a scan shows how something actually is, including wear, later modifications, and deviations from the drawing.

Point cloud or Gaussian splatting?

For a few years now, a second method has stood alongside the classic point cloud: Gaussian splatting. Instead of hard polygon surfaces, it assembles many soft, spatially oriented Gaussian primitives, each carrying its own position, orientation, colour and transparency. The result looks photorealistic and runs remarkably smoothly in the browser, particularly with vegetation, reflections and fine detail — exactly where a mesh regularly struggles.

What matters for telling these apart is what a splat isn't: a measurable, editable model. Edges, surfaces and component dimensions aren't present as planable geometry, and metric accuracy has to be established separately via control points. For surveying, CAD and BIM, the point cloud and mesh remain the foundation.

The practical answer, then: splats are an excellent building block for viewing — say, to show a real installation site quickly and convincingly in the browser. For anything that needs to be processed further, measured, or broken down into components, the route runs through the point cloud.

And what's a mesh?

Between the point cloud and the finished application sits a stage that often goes unmentioned in proposals: the mesh. A point cloud consists of millions of individual measurement points with no relationship to each other. A mesh is created from it by connecting these points into a closed net of triangles. Only then does the object have surfaces that can be coloured, lit and displayed in the browser.

The chain has three parts, then: the scan produces the point cloud, the point cloud becomes a mesh, and the mesh is prepared for the specific application — reduced, structured and given materials. Designed models take this detour too: a CAD dataset describes surfaces mathematically precisely, and for web and AR it's likewise converted into a mesh.

The triangle count matters in practice here. A mesh straight from the scan easily runs to several million triangles, and no browser will render that smoothly. Reducing it to a sensible level without the shape suffering is a distinct piece of work, and it belongs in any serious proposal.

Where a scan is the wrong route

A scan only sees surfaces. It knows nothing about the inside and has no concept of assembly structure. A scanned gearbox can't be opened up, because there's nothing inside the scan at all.

Two limits follow from this, and you should know them before commissioning. A scan is unsuitable for training and explaining internal workings: anyone wanting to show how something functions, which part moves where, or what happens behind the housing needs a designed model. And it's a poor foundation for later expansion: if you only need a view today but want to build a configurator or a training module tomorrow, a surface with no structure won't get you far. That decision gets made at the start, and correcting it later takes considerable effort.

What you need a 3D scan for: existing plant with no CAD data left. Third-party components. Installations that have grown organically in a hall. Surveying an installation situation on site at a customer's. And reconciling design against built reality.

The digital twin: the model that runs alongside

A digital twin is a model permanently connected to the real object. It lives alongside the machine: sensors report operating states to the model, and insights from the model feed back into the real installation.

This connection is the defining feature, not the visuals. Research has drawn a clear line for years. Kritzinger and colleagues distinguished three stages in 2018, differing purely by the degree of data coupling:

• Digital model: no automatic data exchange. Changes to the real object don't automatically affect the digital one, and vice versa.
• Digital shadow: automatic data flow in one direction, from the real object to the digital one. The model knows what the plant is doing but doesn't feed back into it.
• Digital twin: automatic data flow in both directions.

By this definition, a lot of what's called a 'digital twin' today is really a digital model, or at best a digital shadow. That's not a criticism — it's simply a different task with different effort involved. A genuine twin requires sensors, interfaces, data management and ongoing operation. It's a maintenance and production project, not a communications one.

Three questions that settle the case

You don't need a technology consultancy to work out what your case is actually about. Three questions are enough:

• Does the object already exist? If not, only a designed model is an option. There's nothing to scan.
• Do you have design data? If so, preparing that data is almost always the fastest route. If not, a scan is where you start.
• Should the model reflect ongoing operation? Only if you say yes here are you actually talking about a digital twin. For sales, trade shows and training, the answer is almost always no.

We ask exactly these three questions in the first conversation, and in most cases it's clear afterwards what's actually being discussed. After twenty years in industrial communications, you learn pretty quickly whether someone's describing a communications problem or an operations problem. Those are two different projects, and mixing them up is expensive.

What this means for communications and sales

For the vast majority of tasks in marketing, sales, trade shows and training, a prepared 3D model is enough. It's the common denominator of every further application: the same dataset yields the web presentation, the digital catalogue, the configurator, the AR view, the trade show exhibit and the training module.

If your company already runs a digital twin in production, that's good news — though not because you could put it to use in marketing. It contains plenty nobody should see in a sales conversation, and little of what would actually convince anyone there. The good news is different: where a twin is running, the underlying data is usually in good shape, and preparing it for communications goes correspondingly faster.

Quick answer

The terminology question is the second step. The first is where in your sales, explanation or service process understanding actually breaks down. That's exactly where we start: we come from B2B marketing for products that need explaining, not from 3D production. So a project with us doesn't begin with a format — it begins with a process. Whether that turns into a 3D model, a 3D scan, a digital twin or something else follows from the specific use case.

If you'd like to talk this through for your product, feel free to arrange a short first conversation at info@hello-flame.com.