From broad segments to shopping state.

The homepage had competing jobs.

The homepage had to balance launches, offers, priority brands, category discovery, replenishment, and newness — while still being useful to the individual user.

That made personalization harder than ranking products. A campaign could matter commercially but be wrong for the customer. A discount could be timely but still not relevant. A trending product could create confidence for one user and hesitation for another, depending on skin type, shade comfort, routine stage, price sensitivity, or trust.

The work was not to add another recommendation module. It was to decide what deserved the first screen for each user: a routine gap, a concern-led prompt, a replenishment nudge, a shade-safe product, a city or weather-led need, a relevant offer, a trusted brand, or proof from similar users.

The first screen also collected signal.

For You needed profile depth to work well. Without skin, hair, tone, preference, or behavior signals, the app could not deliver a meaningfully different homepage.

So new and low-signal users saw lightweight prompts instead of weak recommendations: answer one skin question, choose a hair concern, confirm shade comfort, pick a routine goal, or share a category preference. Each input unlocked better cards over time.

The principle was simple: do not pretend to personalize without signal. Use the first screen to earn the next useful signal, then turn that signal into a more relevant experience.

Every eligible card had to compete for the first screen.

We prioritized cards across five questions: is it relevant to the user, is the signal strong enough, is the timing important, does it support a business priority, and has the user already seen or acted on something similar?

Time-sensitive cards like price drops, back-in-stock alerts, cart recovery, and replenishment could win when the intent was clear. Routine gaps and concern-led prompts won when they solved a specific next step. Twin-backed cards only appeared when there was enough similar-user density.

Each card started with a user state, not a marketing message.

We mapped common shopping states to card types. The copy came last. First came the signal, the product logic, and the reason the card deserved homepage priority.