In vertical fashion, design decisions happen long before a product reaches stores. Colors must be committed to, sourced, manufactured, shipped, months or even years ahead of a trend becoming obvious. The standard creative approach relies on designer instinct, but instinct is hard to scale, defend to stakeholders, or time precisely. Lululemon needed a way to surface early signals that could either validate creative hunches or catch emerging colors before competitors did.
“We need to make strategic bets on things like color before many people realize. Understanding the precursors to trends, what culture is saying about things that may happen in the near future, is really interesting to us.”
— Ryan Fry, Director of Product Analytics & Data Science, Lululemon
The approach
Lululemon partnered with Nichefire, a cultural foresight platform, to build a two-engine color detection system. Rather than tracking fashion directly, the model listened upstream, to the cultural arenas where colors get tested before they reach apparel.
Engine 1.0 — historical validation: Analyzed 20 months of data and 170,000+ social mentions to identify which behavioral patterns preceded durable color adoption in the past.
Engine 2.0 — live detection: Monitored upstream cultural arenas in near real-time, beauty, film and TV production design, sneakers and streetwear, and broad color discourse — to catch emerging signals before they hit fashion.
Triangulated signals: Each color was scored across three dimensions: discourse (what people are saying), intent (what they’re actively searching), and influence (what media is amplifying).
The logic: when the same behavioral patterns that historically preceded color adoption begin reappearing in live upstream data, confidence in the color’s future commercial viability rises significantly.
The five adoption behaviors tracked
Low-risk trial — color appears first in nails, accessories, and home decor, where commitment is cheap and experimentation is easy.
Cross-context spread — the color starts showing up across multiple unrelated categories simultaneously.
Emotional language — people begin attaching feelings to the color (“it feels grounding,” “it feels warm”), rather than simply naming it.
Neutral behavior — the color becomes easy to coordinate with existing wardrobes, signaling staying power over trend-of-the-moment status.
Shade specificity — consumers stop saying “blue” and start saying “periwinkle” or “ink blue,” indicating deepening conviction.
High-confidence color families identified
Burgundy / oxblood
Sage / olive / smoky jade
Cognac / chocolate / mocha
Taupe / soft stone / sand
Periwinkle (emerging)
Milky pink / pale blush
Each family was scored out of seven adoption behaviors. None hit seven out of seven, which is by design; the model is intended to surface colors early, before they peak.
Results and validation
Early signs suggest the model is tracking well. Cobalt blue was flagged by the engine and subsequently trended in line with expectations. Butter yellow surfaced in an automated report to a Lululemon team member who was already independently tracking it, a clean cross-validation of the signal. The nail art hypothesis was also borne out, with engine 1.0 consistently surfacing nail trends as leading indicators of future fashion colors. Perhaps the most telling example was milky pink: when the product arrived at a hand-curated store, staff were skeptical. It sold out anyway and became impossible to keep in stock — exactly the kind of outcome the model is designed to anticipate before creative consensus catches up.