When text falls short
There are three situations in which a written instruction falls short on principle, however well it is written.
The process happens where you cannot look: inside a pipe, inside a reactor, below ground. It turns on spatial relationships you have to see to understand. Or the sequence is complex and mistakes in it have consequences that only surface later.
In those cases, imagery is not a decoration on the instruction. It is the instruction.
The forms and where they belong
Technical illustration and exploded views. The cheapest step up. Excellent for assembly sequences and for naming components. Static, so unsuited to processes that develop over time.
2D process animation. Explains what happens inside the system: where the flow goes, where the pressure builds, what happens when a valve fails. Relatively affordable and highly effective for understanding, not for operations.
3D work instruction. The operation itself, step by step, with the ability to rotate and zoom in. This is what an engineer uses at the machine. More expensive than animation because a usable model has to exist, but reusable across many instructions.
Digital twin. A virtual representation of an existing installation, often fed with live data. Intended for management, analysis and planning; training is a by-product, not the goal. Heavy and costly, and only defensible when there are operational reasons too.
VR simulation. Think of it as 3D plus interaction plus consequences. The most expensive form and the narrowest application: situations you cannot rehearse in reality.
The cost question
The most important cost variable gets skipped almost every time: does a usable 3D model already exist?
On new build there is usually a design model. That saves a great deal, provided it is usable for visualisation; engineering models are often too heavy and too detailed and have to be simplified. On existing installations from the nineteen-eighties there is nothing, and it begins with measuring or scanning.
So ask that early. The answer moves the budget more than the choice between animation and 3D does.
What you reuse
The argument that makes visualisation defensible sits here, and it rarely gets made.
A model you build once then feeds all sorts of things: the work instruction on the tablet, the animation in the e-learning, the imagery in the manual, the illustration in the quotation, and, if it comes to that, the VR scenario. The first application carries the full cost; every one after it is a fraction.
That also means the first choice weighs more heavily than it looks. Whoever has the model built as a one-off part of a single instruction pays the full price for one application. Whoever fixes up front that the model is their property and gets delivered reusable is buying a foundation rather than a product.
Ownership of source files applies to all your learning material, not only to 3D. Why that matters so much is in the guide on setting up or renewing a corporate academy. What it delivers in practice is under 3D work instructions.