3D product configurators that answer the questions your sales team keeps answering

What this actually solves

A manufacturer of anything configurable is quietly paying for the same conversation over and over. What does this look like in walnut. Can I get it with brass legs. Does the three-seater come in that fabric. Every one of those questions costs a salesperson's time, and the ones that arrive from a different time zone sit unanswered until your office opens. For a made-to-order business the alternative has historically been physical samples and photography — workshop time and a photoshoot to show one combination out of everything the catalogue can build, after which the buyer asks to see a darker one.

The deeper problem is that the buyer cannot commit to what they cannot see. Send four photographs and say the other finishes will look similar, and a serious order gets smaller, because trust does not stretch that far when someone is signing for a container. And after the sale there is the expensive half of the same problem — returns. A customer who has turned the product around, looked underneath it and chosen the fabric themselves sends it back far less often, and in furniture, fittings and equipment a return is expensive to absorb — freight in both directions, handling, and an item that may not go out again at full price.

A configurator takes those questions off your team and hands them to the customer, at two in the morning, in their own currency and their own language, on their own phone. What comes back is not an enquiry saying "interested in the dining set" but a fully specified request with the finish, the size, the fabric and the price already fixed, which your sales team only has to say yes to.

How the work runs

1. Map the options before anything is built. A structured sheet of every option group, every choice inside it, what each choice costs, and every combination that is not physically possible — a narrow cabinet that cannot take double doors, a fabric only offered on certain frames. You fill it in and sign it off. This document is the project. It decides the price, the schedule and the data model, and starting without it is how configurator projects go wrong. Three to five days.
2. Audit the asset and write the budget. Your model gets opened and three numbers recorded — triangles, separate objects and materials — against what the web actually needs — tens of thousands of triangles rather than millions, a handful of objects rather than hundreds, and a payload measured in single megabytes. Alongside it a written performance budget naming the target phone, load time, frame rate and texture ceilings. Both go to you before work starts. Two to three days.
3. Prepare the model. Invisible geometry deleted, parts joined by material to cut draw calls, decimation taken to the point just before the silhouette suffers, UVs unwrapped, scale and origin fixed, and every swappable part separated into its own object. Materials get named in a readable convention — the wood, the seat fabric, the metal leg — because the code finds them by name at runtime, and a model that exports its materials as Material.001 and Material.002 produces code nobody can maintain. Then compression through the gltf-transform CLI with Draco and KTX2 textures. Two to five weeks.
4. Build the application. React Three Fiber for the scene, a state layer holding the current configuration, the rules engine enforcing the impossible combinations, live price assembly, and a shareable URL that encodes the whole configuration so a buyer can send it to a colleague. Textures are loaded on demand rather than all at once, so choosing the fourteenth fabric does not mean downloading all fourteen up front. Two to five weeks.
5. Connect it to the commercial end. Every combination resolves to a code your workshop or ERP can act on, the configuration goes to the cart or into a quote request, and a downloadable specification sheet is generated from the canvas with the exact combination, dimensions and price — the document your buyer forwards to their own client. One to two weeks.
6. Test on real devices, then hand over. A real phone over USB debugging for genuine frame rates, plus the options JSON, a short document on adding a new finish yourself, and analytics wired to record which options people actually pick. One to two weeks.

What moves the price

The range is $3,000–$30,000 over 6–16 weeks, and three factors decide it.

How many products. One product with several finishes and fifty products sharing a configurator are different projects wearing the same name. Each additional product carries its own model preparation, its own option rules and its own testing.

Whether it reaches an order. A viewer that shows combinations is one thing. A configurator whose output becomes a cart line, an SKU and a manufacturing specification is another, because now it sits in the path of somebody's money and has to be correct every time, not merely convincing.

Texture swaps versus geometry swaps. Changing a colour or a fabric is a texture change and is cheap. Changing a size, a leg style or a door count is a different mesh that has to be modelled, prepared and compressed on its own. Adding sizes can cost more than adding fabrics, even when the fabric list is far longer.

What usually goes wrong

The options get written into the code. In the first version, a condition checking whether the fabric is velvet feels like the obvious thing to type. Then six new fabrics arrive, then eight more, then a rule that velvet is only available on the three-seater, and the code becomes a maze that nobody dares touch. Driven from a JSON file from day one — even when there are only two options — adding a finish stays a data change forever.

The quote is written before anyone opens the CAD file. Priced as a coding job, the project then spends its first fortnight inside Blender deleting bolts that will never be seen, and those hours were never billed. Asset preparation belongs on the quote as its own line with its own hours — your file was drawn for a factory and not for a browser, and making it browser-ready is a job in itself.

The combination count gets confused with the build count. Told there are hundreds of combinations, people either price a hundred models that were never needed, or promise a fortnight for what is genuinely a month of asset work. Combinations are generated. Assets are built. The quote has to be written against the second number.