53% intelligence-ready across this cohort
System state first, agent configuration second — the real pipeline behind the Pennsylvania beverage-intelligence pilot.
Restaurants flagged for human judgment, clustered by shared root cause — one decision can resolve several at once.
Open Judgment QueuePaused mid-run for a real environment reason (e.g. closed for the night) — resuming automatically from the last completed stage. Closed isn't failed.
Confirms each location has functional online ordering, excluding static menus. Real accuracy: 74% initial, retrained to 100% via full human validation.
From this pilot's real per-restaurant data (100% QA sample).
Identifies the ordering system (Toast, Square, Slice, 50+ platforms) via checkout URL/DOM signature matching.
From this pilot's real per-restaurant data (5% cross-validated QA sample).
Captures every listed beverage — name, brand, sub-brand, price — including generic listings that resolve to a real brand via item variations.
From this pilot's real per-restaurant data (7% QA sample).
Adds an item to cart and proceeds through checkout to see if a beverage upsell is prompted. Enterprise comparison: ~45%.
Reported in the pilot spec's own broader sample — not run on this 5,279-restaurant cohort yet.
Identifies meal bundles containing beverages; of those, ~46% include Coca-Cola, ~36% PepsiCo, ~32% generic beverages only.
Reported in the pilot spec's own broader sample — not run on this 5,279-restaurant cohort yet.
Evaluates product images for presence, type, and content, across ~57,800 beverage products listed pilot-wide.
Reported in the pilot spec's own broader sample — not run on this 5,279-restaurant cohort yet.