The dock, the yard, and the pick line
Most warehouse loss-prevention budgets get spent in the wrong order. The stockroom is what people imagine when they hear "theft," so that's where the cameras land first. The actual shrink leaks at the dock, the yard, and the pick line. Drivers and dock workers in the gap between the seal log and the camera record. Trailer swaps and plate mismatches in the yard. Small-quantity high-value picks that don't reconcile until the quarterly count, when the trail is cold. AI cameras don't change where the loss happens. They change whether you catch it in real time or three months later on the P&L.
What changes at the dock
Edge-AI cameras on every dock door classify at the point of capture: which trailer is in which door, when the seal was applied or broken, what the worker did during the load. Modern VMS platforms write a video clip into the WMS record for every receive and ship event automatically. When the seal log doesn't match the trailer, the dock supervisor's phone fires an alert before the trailer leaves the yard.
License-plate recognition at the gate (Verkada LPR, Avigilon ALPR, Eagle Eye, Hanwha) ties every truck arrival and departure to a plate, timestamp, and clip. When the BOL says one trailer and the gate camera saw another, the discrepancy is queryable in seconds. For C-TPAT-certified facilities, this is also the chain-of-custody evidence CBP wants during a Supply Chain Security Specialist site visit.
What changes in the yard
Yards are where most shrink-detection programs are blindest. Trailer pools, drop-and-hook, and after-hours pickups happen with limited supervision. Perimeter cameras with person-and-vehicle classification, motion-zone alerts, and after-hours intrusion detection cover the gap, with gate LPR mapping every vehicle to a record. After-hours alerts route to a UL-listed central station for verified-alarm dispatch, which produces visible police response and shifts the deterrent calculus.
For yards over a few acres, panoramic cameras (Avigilon multi-sensor, Hanwha PNM-series, Axis Q-line) cut camera count without losing coverage. The math usually favors fewer high-resolution panoramic cameras over many fixed ones across a yard footprint.
What changes at the pick line
Pick-line cameras tied to the WMS turn footage into transactional evidence. Every scan event writes a clip into the WMS record, so a variance at quarterly count becomes a queryable list: SKU, date range, person, dock door. Investigation time drops from days to minutes, and the deterrent effect on workers who know the cameras are watching cuts at-the-source shrink. That combination produces the 12-to-18-month payback loss-prevention managers measure.
Camera-agnostic analytics (Intenseye, Dragonfruit AI) add forklift-proximity and slip-and-fall detection on the same fleet, also capturing pallet-damage events that show up as freight claims and insurance increases rather than P&L theft. Both reduce together when the camera workflow is wired to the operational system.
What it costs and where the payback comes from
Public benchmarks for a mid-size warehouse or distribution center (50,000 to 250,000 sq ft) land at $75K to $350K turnkey, midpoint near $175K (SDM Magazine 2025 Industry Forecast Study). Per-camera installed cost runs $1,800 to $3,500. Larger multi-building campuses (250,000+ sq ft) run $250K to $1.2M. Camera count, cabling complexity, high-bay ceiling height, and reusable infrastructure drive the variance.
Payback comes from three places: direct shrink reduction, fewer freight-claim disputes, and a lower insurance premium at the next renewal. Carriers adjust premium downward when camera coverage and verified-alarm response are documented, and that third source is often underestimated. See forklift safety analytics and the warehousing security overview.