The brand had strong products and steady traffic, but under 1% of visitors converted. Watching the funnel made the reason plain: shoppers browsed foundations and concealers, hovered over swatch photos, and left. A tiny square of colour on a screen could not tell anyone with real skin, real undertones and real lighting whether a shade would suit them, and shoppers knew it.

The shoppers who did push through and guess often guessed wrong. Shade mistakes sent nearly a quarter of foundation orders back as returns, which meant the cost of every misjudged swatch was paid twice: once in the lost customer, once in the reverse logistics. Each return also eroded the confidence of a shopper the brand had paid handsomely to acquire.

That acquisition cost was the sharpest edge of the problem. Paid traffic kept rising in price while the funnel quietly leaked at the same decision point, so every campaign poured more expensive visitors onto the same broken step. The brand did not have a traffic problem or a product problem: it had one unanswered question standing between browsing and buying.

We ran a five-month CRO programme built around a single conviction: solve the shade decision and the rest of the funnel would follow. The centrepiece is a guided shade-match quiz that asks photography-calibrated questions about undertone, replacing the guesswork of squinting at swatches with a short, structured conversation about the shopper's own skin.

Behind the quiz, a recommendation engine maps each set of answers to specific foundation and concealer shades, and a redesigned product page carries that match through to checkout, so the confidence built in the quiz never drains away at the final step. Nothing shipped on opinion: every change went out behind sequential A/B tests with holdout traffic, so we always knew what the site would have done without it.

Session recordings and funnel analytics steered each iteration. When the data showed shoppers stalling inside the quiz itself, we trimmed the flow from 19 steps to 12, and completion improved from 44% to 68%. The programme worked as a loop: observe where shoppers hesitate, form a hypothesis, test it against a control, keep only what wins.