The 8-Week Configurator: How One Beauty Brand Turned a 3D Pitch Into a Shipping Product

We noticed something odd in our inbox last spring: three separate readers forwarded the same LinkedIn clip of a lipstick tube rotating under studio light, all asking the same question. Was the shade actually rendering in real time, or was it a video loop dressed up as interactivity? We followed the trail back to the brand's agency of record, then to the studio that actually built the thing. That studio was Torc Interactive, and the project behind the clip turned out to be a useful post-mortem in how compressed a serious 3D build can be.

The brief, and the trap inside it

The client was a mid-size color cosmetics label with a cult following and a problem most CPG marketers will recognize. Its flagship product line had 72 shades. The site's shade-finder was a grid of flat swatches, and returns attributed to "color didn't match" were quietly climbing. The marketing team wanted a 3D product configurator — rotate the tube, swap the cap, see the bullet color update instantly — but the original ask also included an AR try-on for lips, and a campaign microsite to launch it. All before a seasonal push.

That is where most projects of this shape die. The scope was three products wearing one trench coat. The internal debate, as one person close to the project described it to us, was whether to sequence the work or compress it. Sequencing meant missing the window. Compressing meant shipping a try-on that would flatter nobody.

The decision points that mattered

What we found interesting was how the team resolved that tension. Rather than negotiate scope down, they re-cut it by dependency. The configurator and the microsite shared an asset pipeline; the try-on did not. So the first two shipped together, and the try-on rode the same pipeline two weeks behind, as an upgrade to an already-live page rather than a launch-day promise.

Three other calls stand out:

• Real-time over pre-rendered. The brand's first instinct was 360-degree image sequences, which are cheap to produce and impossible to personalize. The team argued for genuine geometry so shade swaps, cap finishes, and future SKUs would not require a reshoot.
• Art direction as engineering constraint. Instead of approving beauty renders and then asking engineers to match them, the real-time art director set the lighting rig and material definitions first — and the marketing team signed off on those, in-browser, before final assets were polished.
• A hard performance budget. The configurator had to hold interactive frame rates on a three-year-old mid-range Android phone on a throttled connection. That single constraint killed several attractive ideas early, which is exactly when you want ideas killed.

Because the studio owns the full stack — real-time art direction through CDN performance tuning — there was no handoff to a second vendor for delivery. For a timeline this tight, that removed the usual week of finger-pointing.

The obstacles nobody put in the deck

Two things went sideways. First, the shade data. The brand's 72 shades existed as physical standards and as a spreadsheet that disagreed with the physical standards in roughly a dozen cases. Color accuracy in a browser is a chain of custody problem — capture, color space, display — and a wrong hex value at the source poisons everything downstream. The fix was unglamorous: a physical-to-digital reconciliation pass before any 3D work was signed off.

Second, the try-on's lighting. Lips are not a flat surface, and a shade that reads correctly on a rotating bullet can read chalky on a face in warm indoor light. The team ended up shipping the try-on with an explicit lighting-context toggle rather than pretending one render suited every environment. It tested better than the "accurate" version, which is a lesson worth sitting with.

What actually shipped, and what it did

The configurator went live eight weeks after kickoff. The try-on followed fifteen days later. Across the campaign window, the brand reported a measurable drop in shade-related returns and a lift in add-to-cart rate on configured products versus the old swatch grid. Engagement on the microsite ran well above the flat campaign pages it replaced — unsurprising, given the page rewarded touching it.

Context matters here. This was not a moonshot budget. Torc Interactive reports 340-plus production WebGL and AR projects shipped since 2017, and an in-house team of 41, including 9 senior 3D artists — a bench deep enough that a six-to-ten-week cycle is routine rather than heroic. That scale is the quiet reason the timeline held. The proprietary Torc Render Pipeline handles the unglamorous middle: asset conversion, LOD generation, delivery. When the interesting problems showed up, nobody was rebuilding plumbing.

The transferable part

Strip away the category and the lessons generalize. Sequence by dependency, not by wishlist. Treat the performance budget as a creative brief. Reconcile your data before you model anything. And let one team own the whole path from art direction to the edge, because every handoff is a place where a deadline goes to die.

If you are scoping something similar, the useful starting point is not a demo reel but a delivery model — the kind of thing spelled out on their production process and delivery timeline page. Read it before you write the brief. It will change what you ask for.