Operational Logistics SOPs

Operations is the test of the cognition. A model that cannot survive contact with a Friday-evening rush in a halal supermarket is not a model.

Branded mesh stores receive a Kolshee Edge Kit: smart shelf labels, RFID/weight-based shelf sensors, a hardened POS bridge, a refrigeration telemetry hub, and a back-of-store fulfillment station. The fulfillment station displays AI-prioritized pick lists for inbound consumer orders, sorted by aisle traversal cost.

• ▸T-0: Order ingested from app, restaurant, or B2B partner.
• ▸T+5s: Inventory check against live shelf signal; substitution graph consulted on any soft-out item.
• ▸T+15s: Pick list emitted to fulfillment station with aisle path.
• ▸T+8m (median): Pick complete, bagged, weighed, sealed, posted to dispatch queue.
• ▸T+25m (median): Customer hand-off via Kolshee batched courier or store pickup.

Kolshee never dispatches a single order solo when batching is feasible. The dispatch optimizer aggregates orders across nearby mesh stores into smart batches, reducing per-order delivery cost by 38–62% versus DoorDash unit economics.

Cross-border lanes are operated as standing programs, not one-off shipments. Each lane has a primary supplier, a hot backup, and a cold backup, with FX hedging programs and customs paperwork pre-templated.

• ▸Daily lane health check by Lane-Health model — risk score, port congestion, FX delta.
• ▸Pre-positioning: high-velocity SKUs pre-staged at U.S. cold storage 3–5 days before forecasted shelf-out.
• ▸Hot-backup activation when primary supplier reliability drops below 0.83 over 14-day window.
• ▸Cold-backup activation on confirmed disruption (port closure, conflict event, customs hold >72h).

Every exception (stockout, late shipment, refrigeration fault, customs hold, supplier failure) is auto-classified by severity, routed to the responsible organ, and escalated on a defined SLA. The Merchant-LLM handles long-tail merchant exceptions in natural language; severity-3+ events page on-call humans.