The engine.

The pipeline

Vatic listens to four public signal streams: web (sites, menus, listings), social (Instagram, Facebook, TikTok), reviews (Google, Yelp, platform-native), and paid search (ad creatives, keyword overlap). Each stream runs on its own cadence: visual snapshots refresh daily, review feeds hourly, paid search and keyword positions every four hours.

The raw intake runs through three normalization layers before it reaches the intelligence engine. Deduplication catches the same competitor change surfacing across multiple places. Source-weighting scales single-tweet noise against multi-platform patterns. Entity resolution maps variations: store names, franchise structures, rebrands: back to the right competitor record.

Confidence, scored

Every insight carries a confidence band: High, Medium, or Directional. The band reflects three factors combined: evidence volume, model agreement across vision/text/change-detection layers, and recency.

• High: strong across multiple channels, consistent, recent. Act on it.
• Medium: real but narrower or still stabilizing. Watch it.
• Directional: early read. One strong signal, or consistent but thin evidence. Get ahead of the move, don't commit to it.

Learning + limitations

Vatic's intelligence sharpens with use: market-specific signal weights, user-specific briefing adaptation. Two to four weeks to calibration. What stays invisible: private operational data, offline moves, and anything outside platform cadence. If it's publicly observable, we'll catch it. If it matters, we'll score it.