3D Configurator

35+ live UK retail deployments. +18% average conversion lift per SKU. 5,000+ e-commerce transactions powered per month at peak. A real-time WebGL configurator platform for any commerce site that needs to put a real product in the customer’s hands before they buy. Open the live REM scene at the bottom of this page to try it.

This case study covers two separate, independently commissioned product-configurator projects from different decades of my work. Section 1 is the 2017–2019 academic Knowledge Transfer Partnership (KTP) research project undertaken at Manchester Metropolitan University. Section 2 is the 3D Configurator platform built at Praxis Digital from 2021 onwards, now continuing as standalone IP and still in production deployment with UK retailers today. They are presented together here because both fall under “product configurators” in my work history. They are otherwise unconnected work — separate commissions, separate stakeholders, separate intellectual property, separate eras.

Section 1 · KTP product configurator research · 2017–2019

The 2017–2019 project was a UK Knowledge Transfer Partnership (KTP) — a commercial-academic venture between Manchester Metropolitan University and a UK furniture manufacturer. The brief was to produce a 3D product configurator capable of letting the manufacturer’s sales force and end-customers explore the full combinatorial space of products that, in some lines, ran to the thousands of variants per model. Traditional studio photography couldn’t economically cover that variant space; the manufacturer needed a way to visualise any configuration on demand.

The project ran with me holding both the KTP Associate role (the placed researcher embedded inside the manufacturer) and the Principal Investigator role on the academic side. The work was originally scoped as a multi-person effort — typically undertaken by a small departmental team — and ended up being delivered single-handed, with commercial and academic outputs running concurrently. Wearing both hats at once was unusual; the only reason a one-person delivery worked at all is the original innovation that came out of the project itself. Without it, the timeline and the staffing model would not have lined up.

The innovation

The innovation was the combination of three components which, individually, were not novel — but combined, produced render throughput orders of magnitude faster than any traditional pipeline:

• A game engine’s render pipeline — Unreal Engine 4, pushed to its real-time throughput limit of around 120 frames per second on commodity workstation hardware.
• An algorithmic data structure that could decompose a product into its variant components and run the arithmetic across every possible combination in milliseconds.
• A simple state system sitting between the two — a way to describe any configuration mathematically and to limit which combinations were physically or commercially valid.

The state system held the mathematical description of the product. The algorithm took that description and worked through the combinatorial arithmetic in milliseconds. The output of that arithmetic was fed into Unreal Engine 4, which used a visibility-state system to match the arithmetic output — at which point a render was generated and the next combination was pushed into the pipeline. For the system to work end-to-end, all of this had to happen within a 16-millisecond per-frame budget, fast enough to let the engine produce 60+ distinct state renders per second inside the UE4 frame loop. A single traditional offline render of an equivalent combination would take at least 20 minutes — so the new pipeline was roughly five orders of magnitude faster than the baseline it was replacing.

Outcomes

The KTP delivered positive outcomes for every stakeholder. The manufacturer delivered a sales uplift of £880,000 in 2019 attributable to the configurator work. The university received the academic outputs and published research. The funder, Innovate UK, recorded a successful project. The KTP Associate (me) graduated with a Distinction. The project produced three formal academic outputs in total: the Innovate UK final assessment, the KTP Distinction (the funder’s top grade), and an MRes thesis on the public record at MMU’s e-space repository.

Read the MRes thesis on MMU e-space →

Section 2 · 3D Configurator platform · 2021–ongoing

The 3D Configurator is a real-time WebGL product configurator platform originally built at Praxis Digital, with development beginning in 2021 and a production launch in 2022. It continues as standalone IP after Praxis Digital wound down operationally in October 2025, and is still in active deployment with UK retailers today. The platform was designed and engineered from scratch — independent work, separately commissioned, separately owned. A fourth-generation rebuild — extending the platform from 3D into 4D — is currently in development; a live demo will be embedded in this case study card when v4 ships.

The platform’s most prominent public-facing customer was REM UK Salons, a major user of the configurator from January 2022 through to mid-2025, when REM entered administration. The live REM scene is preserved at the bottom of this page for portfolio reference; every other deployment runs as a confidential white-label service, and the identities, scenes, and configurations of the active client cohort are not disclosed.

The platform’s distinguishing feature isn’t the 3D engine itself — those are commodity components by now — but the URL-as-state contract: every configuration the customer dials in becomes a stable, shareable, server-renderable link.

The URL-as-state contract

The platform’s central design choice is to put every aspect of the configurator’s current state — product, materials, camera, lighting — into the URL itself. The implication is that the URL is the configuration, not a pointer to a server-held one. The encoding scheme that makes this compact and resilient is kept private as part of the white-label IP, but the practical effects it produces are visible across the deployment cohort:

• Stateless server side — the configurator app holds no per-customer state. Two visitors with the same URL get exactly the same scene. Visitors are not tracked, identified, or persisted server-side.
• Shareable by default — every URL is a deep-link. The Share button serialises the current state; a QR code does the same for AR or print handoff.
• Search-indexable — each configuration is a unique URL. When the configurator is embedded inside host PDP pages, those parameterised URLs become crawlable surfaces for long-tail product / material combinations.
• Server-renderable — the same URL can resolve to an offline-rendered image of the configured product, so a configuration link inside an email can show a real thumbnail without the customer clicking through.
• Replayable across years — sales conversations are reproducible. A configurator URL from June 2023 still resolves correctly today because nothing about the URL is time-dependent or session-bound.

Tech stack

Real-time WebGL on the render side via Three.js and PlayCanvas; Next.js on the orchestration side. The platform ships as a thin embed shim that drops into any storefront — Shopify, WooCommerce, Umbraco, or custom-built — without imposing constraints on the host site’s tech choices.

Deployment surfaces

The configurator is deployed across five distinct surfaces, all using the same URL contract and the same embed:

• Product detail pages (PDPs) — the primary surface; embedded directly into storefronts as an interactive PDP module.
• Showroom kiosks — full-screen embeds running in-store for hands-on customer configuration.
• Trade-show stands — kiosk-style embeds on large displays for live demonstration.
• Sales teams on the road — mobile-optimised view for field sales on tablets, with the URL of the configured product captured and sent to the customer at the end of the conversation.
• Printed QR codes — brochure, catalogue, and packaging integration; the QR resolves to the configurator at the exact configuration shown in print.

Conversion impact

Across the 35+ retail deployments, the platform produces a measurable lift on PDP performance versus non-configurator equivalents in the same store. The numbers below come from a representative deployment cohort and have held within roughly ±2 percentage points across the wider client base.

v4 — into the fourth dimension

The fourth-generation rebuild is in active development. The aim is to take everything the URL-as-state contract enables in three dimensions and extend it into the time dimension — letting a single URL encode not just what the product looks like, but how it moves, opens, installs, ages, or operates over time. The same URL still resolves to the same scene on every device, but the scene becomes a four-dimensional experience rather than a static three-dimensional one. A working demo will be embedded in this case study card when v4 ships.