Short answer
Many fulfillment sites already record pack stations to settle disputes and verify orders. That footage can support activity recognition, coarse task timing, exception detection and workflow statistics. It rarely shows hand detail, an egocentric viewpoint or manipulation quality, and it raises worker and customer privacy questions. Evaluate coverage, alignment with system events, and recording notices before licensing.
Why pack stations are a common source
Warehouse labor is concentrated in a few repeated tasks. Order picking is described in the literature as the most labor-intensive and costly warehouse activity, with its cost estimated at as much as 55% of total warehouse operating expense (de Koster et al., 2007). Packing sits downstream of picking and is often handled at fixed stations that are easy to camera.
Operators record these stations for reasons unrelated to AI: proving what was packed, handling claims, loss prevention and training. That makes a recording archive an existing asset, which a buyer can evaluate without commissioning anything. The question is whether it is the right asset. For the demonstration side, see packing demonstrations.
What the footage can support - and what it cannot
| Use | Pack-station footage | Notes |
|---|---|---|
| Activity recognition (scan, pick up, pack, seal, label) | Yes | At the station level |
| Coarse task timing and cycle time | Yes | Needs reliable timestamps |
| Exception detection (missing item, wrong box) | Yes, with event labels | Link to order exceptions |
| Workflow statistics and station layout | Yes | Useful for scoping |
| Hand detail and grasp type | Rarely | Overhead cameras at distance |
| Egocentric viewpoint | No | Fixed cameras only |
| Manipulation quality or force | No | Not visible |
| Robot action data | No | Human work, no actions |
| Item-level identification | Partly | Small, occluded, in motion |
Fixed-camera footage supports understanding of the workflow, not training control. The gap is explained in fixed-camera footage limits, and the choice between archive and new recording in existing footage vs commissioned recording.
Checklist: coverage and quality
| Check | Pass looks like | Evidence to request |
|---|---|---|
| Camera view | Station, packer's upper body and work surface in frame | Still frames from each camera position |
| Resolution and frame rate | Items and labels discernible at working speed | Recording settings; a 10-minute sample |
| Hand visibility | Hands visible during item handling for most of the cycle | Reviewer count of sampled frames |
| Time base | Reliable timestamps; known clock source | Timestamp and drift description |
| Continuity | Complete cycles, not event-triggered snippets | Retention policy; sample of raw files |
| Variety | Several stations, shifts, packers and box types | Counts by station, shift and SKU class |
| Coverage of exceptions | Short-packs, rework and damaged items represented | Exception log linked to clips |
| Retention | Footage still exists for the periods offered | Retention schedule, with dates |
Many systems keep video only for a short window or only for flagged orders. Footage selected for disputes over-represents problems and may miss normal work; footage on a rolling window may not exist for the dates quoted.
Checklist: alignment with system events
Video becomes more useful when each clip maps to a record. Ask whether recordings can be matched to order IDs, scan events, carton IDs and pack timestamps from the warehouse management system. With these links, you can label start and end of tasks automatically and attach outcomes such as short-ships, without relabeling every clip. See aligning video with system logs.
Check clock synchronization between the cameras and the system, since drift of even a few seconds breaks event alignment. Check that identifiers are stable and can be pseudonymized for delivery.
Checklist: people, notices and permissions
Pack-station footage records employees working, and sometimes customers' order details. The legal questions depend on location, employment arrangements and notices, and this is not legal advice.
- Notice: what were workers told about recording, for what purpose, and does that purpose include licensing to third parties for model training?
- Impact assessment: regulators have published guidance on video devices and on monitoring workers; the European Data Protection Board's guidelines address personal data processed through video devices (EDPB, 2020), and the UK regulator's guidance recommends a data protection impact assessment for monitoring likely to result in a high risk to workers (ICO, 2023). Suppliers in the United States are subject to different rules, but these documents illustrate what a diligent buyer asks about.
- Customer information: order labels, addresses and screens may be visible.
- Third-party content: product packaging, brand marks and client-owned inventory may belong to the 3PL's customers; see 3PL providers as warehouse data suppliers.
- Redaction effect: blurring faces is possible, but heavy redaction of hands or labels reduces training value, and redaction does not by itself resolve a permission question. See worker consent and identifiable people.
Red flags
- The supplier cannot state the original purpose and notice for the recordings.
- Footage is only available as flagged clips with no sampling of normal work.
- Timestamps are from an unsynchronized camera clock.
- The site's clients own the inventory, and client consent has not been addressed.
- Redaction was applied by blurring entire work surfaces.
- The operator wants to license footage of employees without an employee-facing process.
- The offer promises hand-level detail from an overhead wide-angle camera.
Illustrative sample request
# Illustrative only - synthetic values
request: pack_station_sample_v0
sample: 30 complete pack cycles from each of 3 stations (synthetic)
selection: random, across 2 shifts and normal + exception orders
per_clip_metadata: station_id, camera_id, shift, start_ts, end_ts, order_event_ids (pseudonymized)
frames: original resolution; redacted copy and unredacted view held for audit by supplier
documents: recording notice text, retention schedule, redaction method note
Questions to ask suppliers
- Why was the footage recorded, and what were employees told?
- How many stations, cameras and shifts are covered, and for what dates?
- Are cycles complete or event-triggered clips?
- Can clips be matched to order and scan events, and how accurate are the timestamps?
- What do cameras actually show of the packer's hands and the items?
- What redaction is proposed, and what does it hide?
- Who owns the inventory shown, and has that party agreed to the use?
- Can a representative sample be reviewed before any commitment?
Discuss your requirements
If you are considering existing recordings of warehouse packing, describe the tasks, uses and constraints to SourceX. SourceX investigates whether US businesses hold suitable footage, checks the data and the supplier's licensing permissions, and follows up about potential sources; a request does not guarantee a match. See the humanoid training data guide, the warehouse and fulfillment data hub, and the catalog entries for video recordings and supply chain and logistics data.
Explore sourcing and licensing options with SourceX
Sources
- Guidelines 3/2019 on processing of personal data through video devices — regulator guidance on video devices.
- Employment practices and data protection: monitoring workers — guidance on worker monitoring and impact assessment.
- Design and control of warehouse order picking: A literature review — order picking as the costliest warehouse activity.