Online foundation buying has a high penalty for error. Shade names are inconsistent, photography lies under different lighting, and most people default to rebuying the shade they already know. Adding more content, smarter sorting or a better quiz doesn't fix a data problem. So before any matching logic existed, we built the measurement layer it depended on.
Before a single line of matching logic was written, we staffed and ran a physical measurement operation. Every foundation shade — across 85 brands, 1,275+ shades — was captured through a spectrophotometer: the actual physical colour of the product, not a label or a marketing swatch. The operation ran at 500+ shades a day, under three seconds each, evaluated on DE2000 colour distance rather than visual approximation.
This was the product decision that mattered most. Crowdsourcing shade data, scraping brand descriptions, or using photography would have reproduced the same inconsistency the user was already frustrated by. Measurement made the data objective — and objective data is what let cross-brand comparison mean something.
The pipeline had one job: take something physical and precise, and return something human and effortless. Every step from the spectrophotometer onward was invisible to the user. What they saw was a percentage and a sentence.
The user-facing system had one key design decision: don't ask for information the user doesn't have. No face camera. No undertone quiz. No colour-science vocabulary. Instead, the Shade Finder asked a single question: what foundation do you already wear? That known shade became the anchor. Every cross-brand match was expressed relative to it — same logic, same measurement, human language.
The result was something genuinely different from other matching tools: confidence that felt personal rather than algorithmic, because it started from something the user had already validated in real life.
Once a reference shade was saved, matching became part of shopping itself — not a separate tool. Category listings led with "Best shade matches for you" and surfaced match percentages above the grid. Product pages showed the top cross-brand shades, a confidence score and a plain-language note: "darker + marginally less yellow." The reference shade stayed visible and editable, so the user stayed in control without restarting discovery.
The same infrastructure that powered the Shade Finder powered search, PDPs and listing carousels — one data layer reused everywhere, not rebuilt per surface.
We launched with foundation — the highest-risk complexion category — as the proving ground for an infrastructure meant to extend across the catalog. At launch the operation was fully live for all users across Shade Finder, listing, search and PDP surfaces. Not a beta, not a subset.
Measure the shade. Then make the measurement disappear.