When 3D is the wrong answer

Why bother writing this

Because a project that should not have happened is worse for us than one we never got.

It ships late, because nobody can agree on what the scene is for. It gets cut back halfway. It launches, nobody mentions it again, and eighteen months later it is quietly removed. Meanwhile it is in our portfolio, and the client tells people what it cost.

The other reason is narrower. A studio that says yes to everything is telling you it has no particular reason to be chosen, and we would rather be the one that says no occasionally.

When there is nothing to rotate

The clearest case, and the most common.

3D earns its place when what you sell is spatial, physical, or hard to picture from a description. A machine. A building. A process with stages. Something whose shape carries information.

A consultancy sells judgement. A law firm sells judgement. Most software sells a set of screens. There is no object, and building an abstract one to have something moving on the homepage is decoration that has to be maintained. It photographs well in a case study and does nothing for the person deciding whether to call you.

The test is unglamorous: can you name the thing a visitor would want to turn around? If you have to invent it, that is the answer.

When your visitors are on the wrong hardware

WebGL runs on the graphics processor. That is what makes it fast, and it is also what makes it uneven.

A recent iPhone handles a well-built scene without noticing. A four-year-old mid-range Android, warm, on a poor connection, does not. You can design around it, and we do, but designing around it is work, and if that describes most of your audience you are paying for a feature most people will get the reduced version of.

Worth checking before anyone quotes: open your analytics, look at the device and connection breakdown for the last six months, and look at it by revenue rather than by visits. If the money comes from constrained devices, the answer changes, and it should.

When nobody is going to look after it

This one gets skipped in every conversation and turns up two years later.

A scene depends on browsers and graphics drivers, and those change without asking. A page of text does not care. A scene can start behaving differently after an update nobody at your company will ever hear about.

So somebody has to open it a few times a year on real devices and check. If your organisation has no arrangement for that, either with us or with anyone, then you are buying something that will degrade on a schedule you don’t control. Published upkeep for interactive sites runs $95 to $195 a month on a care plan and well beyond that on an agency retainer. If that line has no home in your budget, the honest answer is not to start.

When the real problem is that nobody understands what you sell

A lot of 3D briefs arrive dressed as a design problem and are actually a message problem.

The symptoms are recognisable. Leads go quiet after visiting. Your salespeople explain the company on the phone in a way the website never manages. Somebody senior says the site "doesn’t feel like us" without being able to say what would.

None of that is fixed by adding dimension. If a visitor can’t tell what you do in the first ten seconds, a rotating object gives them something to look at while remaining confused, and it makes the confusion heavier to download.

Fix the message first. It costs a fraction of a 3D build. And if it turns out the message needs the object to be understood, then you have a real reason for the scene and a brief that will survive contact with a quote.

When speed is the entire proposition

Some sites have one job and it is measured in milliseconds. A checkout. A booking flow. A support page somebody has landed on while annoyed.

Google does not penalise 3D, it has no way of knowing what is inside a canvas. It penalises slow. And the metric that catches a scene is INP, which measures how long the page takes to respond after you interact with it, with a target under 200 milliseconds. A scene compiling shaders on the main thread while somebody is trying to click "pay" produces exactly the failure that metric exists to find.

You can build 3D that keeps those numbers. Our own homepage runs a full scene and ships 186 KB of compressed JavaScript. But on a page whose only purpose is to get somebody through a transaction, the right amount of 3D is usually none, and the argument for it is almost always internal rather than about the visitor.

When it is one page of a much bigger problem

If the site has three hundred pages, a content manager nobody can use, and an information architecture that grew rather than got designed, a spectacular homepage does not help. It raises the contrast between the front door and everything behind it.

People notice that. A visitor who is impressed on arrival and lost two clicks later trusts you less than one who was never impressed.

So when is it right

For balance, because all of the above is the case against.

It is right when the object is the argument. When a visitor understands the product better after turning it around than after reading three paragraphs. When the thing you sell is expensive, considered, and currently represented by photographs that undersell it. When you are asking somebody to imagine a space they cannot visit. When configuration is genuinely part of the purchase and a grid of variant photos is a poor substitute.

And, less often admitted but real: when the category is asleep and looking different is worth money. That is a legitimate reason. It is just a different one from the reasons above, and it should be argued on its own terms rather than dressed as usability.