Should You Build a 3D Product Configurator In-House? A Build vs Buy Guide for Manufacturers

Yes, a manufacturer with a capable web team can build a 3D product configurator in-house: Three.js is open source under the MIT license, and the 3D viewer is the part your team will finish first. Whether you should depends on everything behind the viewer (product rules, pricing, the 3D asset pipeline, the quote and BOM handoff, an admin your sales team can use, and years of operation) and on whether your team will still own that work in three years.

Our founder has built 3D configurators since 2016. Door Builder and Muebles Torga run on Three.js, Pepart on React Three Fiber, and AV Drop and NOMI on ShapeDiver.

What does building a 3D configurator in-house actually involve?

Building a configurator in-house means building seven things, and the 3D viewer is only one of them. Here is each part, and where the effort went in configurators we shipped:

Part	What your team builds	Where the effort went for us
3D viewer	Scene, camera, materials, lighting, option swaps	The smallest share: Three.js does the rendering
Product model and rules	Which options exist, which combinations your plant can build	Door Builder: options follow how the shop floor routes each door
Pricing	Price computed from the configuration, per customer tier	Door Builder: residential or commercial pricing applied in the cart
3D asset pipeline	CAD to web-ready parts, compressed; finishes as materials	New parts, product lines and finish textures, long after launch
Quote, order and BOM handoff	Quote, checkout, order record, bill of materials, ERP export	Pepart: one crop calculation drives the preview and the 300 DPI print file
Admin	Screens where non-developers change options, prices and rules	Every change that would otherwise be a developer ticket
AR, mobile and operation	Phone layout, AR export, monitoring, updates	New product lines, rules and prices, every season

A proof of concept that covers the first row proves the library works. The business case sits in the other six.

Who do we need on the team to build it?

You need five skills, whether they sit in five people or fewer:

• A 3D web developer who knows Three.js or React Three Fiber and phone performance.
• A technical artist who turns CAD into web-ready parts and materials.
• A backend developer for rules, pricing, orders and the store or ERP integration.
• A product owner who knows the rules: what the plant builds and how each option is priced.
• Someone who operates it: monitoring, updates and the next product line.

In what order should we build it?

Build the rules and pricing model first, because every other part reads it; then the viewer, then the quote, order and BOM handoff, then the admin and AR. Our own Diagnose starts the same way: about a week mapping your rules into a spec before anything is built. Drive the 3D viewer from that model instead of giving it its own copy of the options, and test the order handoff against real orders your plant has already built.

Is the 3D viewer the hard part?

No. The 3D viewer is the most visible part of a configurator and the least risky one to build yourself. Three.js describes its aim as an easy-to-use, lightweight, cross-browser, general-purpose 3D library, and it is MIT-licensed. React Three Fiber is a React renderer for three.js whose documentation says everything that works in Three.js works there too, so the choice follows your app stack. Babylon.js, a JavaScript game and rendering engine under the Apache 2.0 license, is an option if your team already knows it.

“General-purpose” is the point: none of these libraries contains a configurator, so the product model, pricing, order and admin are what your team signs up for.

Where do the product rules and pricing live?

In an in-house build, the rules and pricing live in code your team writes and maintains, and in our builds they take more of the work than the viewer does. The rules decide which combinations your plant can build: a hinge that only fits certain door weights, a module that does not fit next to another. Getting them out of engineers’ heads is its own discipline, covered in Product Modeling: Turning Tribal Knowledge into Configuration Rules.

Two design decisions matter more than the 3D library:

• One option model for everything. The 3D scene, the price and the order read the same rules, so the render and the order cannot disagree.
• Pricing as rules, not a list. Configurable products price from drivers such as size, material and finish class, with retail, trade and dealer tiers on top. A flat price table works for a demo and does not hold those tiers.

Getting the rules right pays. Once the Muebles Torga configurator stopped buyers from ordering parts that do not fit together, customer-service tickets fell 72%.

How much 3D asset work does a configurator need?

A configurator needs one web-ready model per part or module, with finishes applied as materials and sizes generated in code; nobody should model one file per combination. Your engineering CAD is the starting point for those parts: it carries manufacturing detail no buyer sees, and as exported it is often too heavy to load on a phone.

The web format is glTF, which Khronos describes as a royalty-free specification for the efficient transmission and loading of 3D scenes and models. Compression is not automatic: the Three.js loader needs a separate decoder for Draco meshes, for KTX2 textures and for Meshopt. After launch, the standing work is new parts, new product lines and a texture set for each new finish.

Run a few of your models through our free GLB optimizer to see what makes them heavy, and read 3D configurator performance for load-time targets.

Do we need AR and mobile support?

You need mobile from day one, and AR as soon as buyers want to see the product in their own space, and AR means two file formats. On Android, Google’s Scene Viewer opens glTF 2.0 and GLB models, and recommends 10 MB or less, textures no larger than 2048 × 2048 and at most 100,000 triangles. The same page lists only two supported glTF extensions, KHR_materials_unlit and KHR_texture_transform, so the Draco- or KTX2-compressed model your web viewer loads is not the file AR opens. On iPhone and iPad, Safari uses AR Quick Look, which displays USDZ files.

The AR model also has to match the buyer’s configuration, so it is a separate export, generated on demand. Open our door configurator on a phone and tap “View in AR”: it exports the door you just configured and opens it in Quick Look on iPhone, or in Scene Viewer on Android after a second tap.

What happens after the buyer clicks “Get a quote”?

After the click, the configuration has to become a quote, an order and, for a manufacturer, a bill of materials and production data, without anyone re-typing it.

Door Builder’s sales team hand-built each quote, matching options to what the shop floor could produce, and a single quote took about two and a half weeks. With option rules that follow how the shop floor routes each door, residential and commercial pricing at checkout, and a dealer portal, a dealer now gets a quote in four minutes. The result came from the rules, pricing and dealer flow behind the 3D view.

If your configurator has to reach production, see how a BOM configurator derives the bill of materials before you scope the build.

Who changes options and prices after launch?

Your sales or product team should, so your build needs an admin for non-developers. If a new finish needs a pull request, the configurator falls behind the catalog. Name a second person who understands the rules code before launch.

When is building in-house the right call?

Building in-house is the right call when a standing web team will own the configurator for years and the product is narrow enough for that team to model completely. The signals:

• An internal web team that already ships and operates production software, with an owner for the rules code.
• One product family, or a few, with rules your engineers can already write down.
• No revenue deadline, so the team can learn the asset pipeline and order handoff as it goes.
• A strategic reason to keep the configurator inside your own engineering organization.

Our Diagnose includes telling you when a SaaS tool or your own team is the better call. If your catalog fits a SaaS product, see our custom vs SaaS comparison and our guide to Threekit alternatives.

When is hiring a builder the right call?

Hiring is the right call when time to revenue matters more than building the capability, or when your rules are complex enough that a first attempt is likely to be rewritten. The signals:

• Revenue or a dealer launch depends on the date.
• Rules depend on other choices, buyers type in sizes, or pricing has several customer tiers.
• The configurator has to produce a quote, an order or a BOM that production accepts.
• Your web team is already booked on the store, the ERP or internal tools.

One signal alone is no reason to hire. Hire when two or more apply and no one on your team has shipped a configurator.

Is there a third path between building and buying?

Yes: hire a builder, then own the result. When we build a configurator, you own the source code and the data, with no license or per-seat fee, and the repository can live in your GitHub or GitLab. It runs on Vercel and Neon in the region you choose, on your own accounts from the start or on ours while we operate it, and we move it to yours whenever you ask.

Your developers can read the code from the first week. We build on Forge, the platform we built for this work (catalog, configurator, checkout, portals and admin; not Autodesk Forge), which already runs Door Builder, Muebles Torga, Nafarú, AV Drop and Pepart, so a build starts from shipped parts. Our Three.js product configurator page shows what that looks like.

Hiring has its own key-person question, and you should ask it of us too: SBFS is one senior engineer working with AI, taking five clients at a time, as our About page says. Our answer is the ownership above. Ask any builder what happens if the vendor disappears, and get the answer in writing before you sign.

How does the cost of hiring compare with building?

Hiring us costs $4,000 one-time for a Pilot, and a build is a one-time setup, from $20,000–$40,000 at Parametric to $50,000 and up for Forge, plus a monthly fee for running and improving it; building costs your team’s time on the seven parts above and on operating them for years, which only you can price. Our tiers:

• Pilot: $4,000 one-time: your own product live in a working 3D configurator on your own site, priced, credited in full to the build if you continue.
• Parametric: $20,000–$40,000 setup and $500–$1,500 a month: any size, files for production, store or ERP.
• Forge: from $50,000 setup and a monthly fee scoped with you: your whole sales operation.

Hosting on Vercel and Neon is billed separately. Diagnose takes about a week, is included and ends with a quote inside your tier and a date; the full channel (the Forge tier) typically ships in six weeks. Details, including the timeline, are on our pricing page.

Frequently asked questions

Can we build a product configurator with Three.js?

Yes. Door Builder and Muebles Torga run Three.js configurators in production. Three.js covers the viewer; the product model, rules, pricing, order handoff and admin are yours to build.

Should we use Three.js or React Three Fiber?

Choose by your application stack. React Three Fiber is a React renderer for three.js, so a React team gets components that react to state without losing Three.js features.

How long does it take to build a 3D configurator?

In-house, it depends on the rules and on whether the team has shipped one before, so we publish no estimate for that. Ours: a free demo of one of your products within one business day, Diagnose in about a week, and the full channel typically in six weeks.

If we hire a builder, who owns the code?

With us, you do: the source code and the data, with no license or per-seat fee. The repository can live in your GitHub or GitLab and run on your own Vercel and Neon accounts.

Sources

• three.js on GitHub: project README and MIT license
• React Three Fiber documentation: Introduction
• Babylon.js on GitHub: README and Apache 2.0 license
• Khronos Group: glTF overview
• three.js documentation: GLTFLoader
• Google ARCore: Scene Viewer
• Apple Developer: AR Quick Look

Before your team writes a prototype, get a benchmark: on a 20-minute call we agree on one of your products, and within one business day of your photos or drawings you get it in a working 3D configurator on a private link. See the free demo and our price ranges, or read how Door Builder’s quotes went from about 2.5 weeks to 4 minutes.