Four layers turn guest language into priced action — on owned hardware, under written consent, with no outbound API call leaving the building.
SPC works only with first-party material a property chooses to provide. No scraping, no shared benchmark feed. Signed consent and a 48-hour onboarding window.
The model reads each review for sentiment, sub-aspect, and urgency, turning raw language into structured hospitality data. Trained on 630,000 NYC reviews across 197 properties; 95.9% label accuracy on held-out data.
SPC cross-references those signals against PMS patterns to estimate timing, booking impact, and revenue at stake — anchored to a per-hotel autoregressive baseline. The 14-day forward signal delivers +0.020 OOS lift, validated on 193 NYC hotels.
It assembles the output into a readable briefing — each line shaped around action, timing, and commercial consequence. One brief per fortnight, served on owned hardware.