AI Intervention, China Export Controls, and Warehouse Automation for SKU Sprawl

AI Moves Exception Management from Review to Intervention

This week, AI pushed deeper into live logistics execution across cargo operations, freight finance, and procurement. SmartKargo added AI apps to its Cargo Control Tower for predictive network visibility, early exception detection, delay and SLA-risk prediction, and disruption decision support, while also extending into ULD tracking, cargo-system and partner messaging, and flight-level capacity and pricing decisions. VietJet Cargo was cited using AIRCAM to improve revenue management, pricing, and capacity response.

In freight billing, AI agents are now reconciling invoices, shipment records, contracted rates, service codes, delivery events, and proof-of-delivery data to flag pricing variances, duplicate invoices, unsupported accessorials, and missing PODs. One deployment at Worldwide Express cut active exceptions 75%, from $15 million to $4 million in six months, closed about 10,000 exceptions weekly with less than 1% error, improved by four weeks, and released $10 million in cash. GS Caltex also deployed AI to automate procurement workflows and detect anomalies faster.

For teams, the work is shifting from chasing discrepancies to supervising AI outputs, validating escalations, and tuning the rules and data behind automated decisions. The advantage will go to practitioners who can govern exceptions, improve data quality, and coordinate across functions, not just process transactions.

China Broadens Export Controls Beyond Drones to Industrial Equipment

China’s Nov. 18, 2026 measures widen export control from drones and components to tunnel-boring machines, walking excavators, high-altitude shotcrete equipment, and dredger pumps, dragheads, and cutterheads, with licenses required for “sensitive” destinations and registration plus onward-transfer limits elsewhere. Beijing’s Oct. 16 draft goes further, adding 40 technologies to the restricted or prohibited export catalogue, including 23 in AI and 12 in photovoltaics. Vietnam adds another checkpoint on Nov. 16, 2026, when Circular 142 begins requiring electronic declarations and recipient identification for certain postal and express imports. The shift is from shipment clearance to shipment feasibility. The question is no longer only whether a product can leave origin; it is whether the product, software, process, installation, upgrade, or re-export can move at all without a license. The U.S. is tightening too, with customs crackdowns on alleged transshipment across 40-plus economies and an Oct. 28, 2026 classification of imported robotics, machinery, and components as a national-security concern. For supply chain and trade professionals, the edge now sits with people who can connect classification, end use, origin, and service scope before orders are booked. Trade compliance, procurement, forwarding, and customs teams need to operate as one control tower, because documentation quality now determines whether revenue clears, not just how fast it moves.

Warehouse Automation Is Now Optimizing for SKU Sprawl

GXO’s Digit deployment at Flowery Branch has now moved more than 100,000 totes at about 98% on-task accuracy, with 65,000-plus cumulative operating hours across customer deployments, and additional rollouts are already scheduled for Statesville and Lebanon in early 2027. Locus also crossed 5 billion warehouse picks, reaching that mark just 24 weeks after 4 billion. The signal here is less about proving robots can work and more about proving they can absorb assortment complexity at scale: Hy-Tek used automated retrieval to manage a jump from roughly 500 to 4,000 SKUs, and First Supply’s automated DC uses 31 robots, three picking stations, and 12 sortation lanes to support branch-specific SKU diversity and lift throughput about 30%. For operators, the next layer of work is no longer commissioning or basic orchestration, but replenishment logic, exception control, and flow balancing across WES/WMS-linked systems. That extends the progression from managed autonomy and digital-twin validation into day-to-day control of highly variable item networks.