the configurator that bored its own founder

Lumen Atelier makes lamps in Kraków. The good kind — second-generation, the founder learned it from his father, the brass gets hand-finished by a guy who's been there longer than the founder's been alive. He sells to interior designers in Berlin, Milan, New York. People who specify, not people who browse. We built him a WebGL configurator. Next.js, Three.js, Payload for the part library, Stripe for the part where money changes hands.

We did the obvious thing first. Of course we did. Catalogue up front, browse the range, find a lamp you like, then a "customise" button drops you into the configurator. It's how every product site on earth works. Browse, then build. We had it running in under three weeks and it looked clean and we were a little proud of it.

Then the founder put it in front of a friend. Not a real usability test — he just opened it on his laptop at the workshop and handed it over while we watched on a screen share. The friend clicked into the catalogue, scrolled, opened a lamp, looked at it, and about thirty seconds in, you could see it. The disengage. The little exhale. He was being polite. He was bored.

Thirty seconds. A guy who likes the founder, looking at lamps the founder's family has made for forty years, and he checked out in half a minute.

That's the moment. Not a meeting, not a deck, not a month of us having opinions about information architecture. One bored friend on a workshop laptop told us more than the entire build had.

So we tore it down. Killed the catalogue as the entry point — demoted it to a backup, a place to land if you don't know what you want yet. The configurator became the front door. You don't browse a finished lamp anymore. You start with nothing and you compose one: pick the arm, the shade, the brass finish, the bulb temperature, and the thing renders in real time while the price counts up beside it. You're not shopping. You're making. And because you're three choices deep before you've had a chance to get bored, you're committed before you know it.

Forcing the commitment is the whole trick. The conventional IA gives people an exit at every step — see a lamp, don't love it, leave. Composition gives them a sunk cost in the best way: that's your lamp taking shape, you picked every part of it, of course you want to see it finished.

Average session went from about 41 seconds to roughly 9 minutes.

Here's the honest lesson, the one that stung a little. The "right" architecture — the one every best-practices article would've told us to build, the one we did build — was wrong for this product and this buyer. Browse-then-build is correct when people know roughly what they want. It's exactly backwards when the joy is the building. We'd have argued about that in the abstract for a month and lost. A thirty-second test settled it for free.

Now we test ugly and early, before anything's polished enough to feel precious. Cheaper to find out a friend's bored on a Tuesday than to find out a market is, six weeks after launch.

see the case study →