Prototype · Emerging-brands radar
Prototype · Emerging-brands radar
You gave us eight sectors. We grew them into a 557-brand universe, scored every brand against its own 9-week baseline across 416.9 billion interactions of social data, then an AI analyst read the underlying posts, threw out the noise, and named exactly what is driving each real surge.
Each sector from your email became a self-expanding universe: household incumbents for baseline, plus every challenger the social firehose says people are actually talking about. Tap a sector to see its full signal board — every brand ranked by emergence score, highest to lowest, with the verified calls colored in.
Momentum screens drown in noise: month names trend every month, common-word brands pollute their own keywords, and one viral post looks identical to a real breakout. This pipeline kills each failure mode in order — quantitative thresholds first, then an AI analyst who actually reads the posts.
Every brand's last 4 weeks vs its own 9-week baseline — growth must show up in interactions and post volume and unique creators. One viral post can't fake all three. A consistency score separates sustained emergence from a one-day spike.
A volume floor (≥20 posts and ≥30K interactions/day) kills tiny-topic noise where ratios explode off nothing. An at-scale ceiling routes household names to their own incumbent board, so a Nike promo can never bury a genuinely new brand.
Every finalist gets its top posts pulled and read: is this conversation actually the brand? What specifically is driving it — a launch, a collab, a creator moment, a controversy? Pure-momentum false positives get named and killed here, with receipts.
Growth against consistency, sized by current daily conversation. The upper-right is where you want to be looking: big, sustained, and accelerating. Hollow red dots are the ones a naive momentum screen would have handed you as "signals" — the verification pass read the posts and rejected them.
Each card is one brand's story: the momentum numbers that flagged it, the 13-week shape of its conversation, who is driving it, and the receipts — the actual highest-engagement posts from the window. Verdicts are earned, not assumed.
Your pipeline already receives the raw n-grams. This is the mapping layer on top. For every confirmed emerger, the agent mines its post pool for every spelling, hashtag, and phrase the conversation actually runs on, validates that each keyword's own posts are mostly about the brand, and only then counts it toward emergence. Spelling variants are caught automatically — Cecred's conversation runs heavily on "cécred" with the accent, a keyword an exact-match screen never finds.
| Brand | Sector | Keywords the conversation runs on · % of posts | Coverage | Brand-pure | Growth |
|---|
The hardest problem in emerging-keyword detection isn't finding spikes — it's that most spikes are lies. These passed every quantitative threshold. Then the analyst read the posts. This is the step that separates a tradable signal from a thousand-keyword dump.
Everything above was generated from the live LunarCrush firehose plus an AI analyst loop. It can run on any universe, any cadence.
This exact analysis, delivered monthly per sector: who's new on the board, who's sustaining, who faded. Same verification bar — every name comes with the driver and the receipts.
Point the pipeline at any category, any watchlist, public or private names. The universe expands itself as verification surfaces adjacent brands people are actually talking about.
Same scoring on a weekly cadence with a persistent board: entries, exits, rank moves. New confirmed emerger = a flag in your inbox with the narrative attached, not a keyword dump.
The product version: point-in-time-aware entity recognition across the full firehose that groups every spelling, nickname, and product into its brand automatically — the step that removes keyword lists entirely. Experimental API first; early access as promised.