The richest data in most facilities is already being produced by people scanning things. Using it well is cheaper than any new sensor.
What a scan event contains
Time, to the second.
Location, if the read point is fixed or the device reports position.
Item and quantity.
Task and order reference.
Device and, usually, operator identity — which is the point at which process data becomes personal data.
Together these bracket every handling step, and the gaps between them are where the questions are.
What the gaps mean
A gap between two scans is unexplained time.
It may be travel, searching, waiting, an interruption, or the work itself, and the scan data cannot distinguish them.
Compare gaps against the distance implied by the two locations. A gap much longer than travel would require is search or wait time.
Consistently long gaps at a location indicate a slotting, labelling or congestion problem there.
This inference is the main analytical technique available without positioning hardware.
Scan discipline problems
Every facility has these and they corrupt the analysis quietly.
Batch scanning. Several items scanned at once at the end of a sequence, so all timestamps cluster and the intermediate times are lost.
Scanning at the desk rather than at the location, which destroys the location signal.
Pre-scanning, where a label is read before the physical action.
Skipped scans with a manual entry afterwards.
Duplicate reads, where the same tag is read repeatedly by a fixed reader.
Detect these by looking for implausible patterns: several scans within a second, sequences with no travel time, locations that could not have been reached in the gap.
Fixing discipline
Understand why first. Batch scanning is usually a response to a device that is slow, a process that requires holding items, or a scan step that adds no visible value.
Fix the cause, which is frequently ergonomic or technical rather than behavioural.
Make the scan useful to the person doing it, with confirmation they need.
Do not attach discipline to scan compliance, which produces compliant-looking data rather than accurate data.
Report the anomaly rate as a data quality measure, trended, and treat a rise as a process signal.
RFID against barcode
Barcode requires a deliberate act, which produces a clean event with a known actor and a known moment.
RFID reads passively, which produces coverage without effort and much noisier data: repeated reads, reads from adjacent zones, reads of things not being handled.
RFID answers "where is it"; barcode answers "what happened".
Both need filtering. RFID needs dwell logic to distinguish a passing tag from a stationary one; barcode needs deduplication.
Making it analysable
One event table with a consistent schema across read points.
Clocks synchronised, which is the single most common cause of nonsense results.
A stable location scheme, since facilities accumulate several historically.
Retain raw events, not just the aggregates, because the questions change and the raw data cannot be reconstructed.
Define the identity policy at ingestion, which is where the process-or-personal-data decision is actually made.
Detecting batch scanning
The most common data defect and it is visible in a single query.
Count scans occurring within the same second, by device and operator.
Look for sequences with no travel time between locations that are physically apart.
Look for clusters at the end of a task rather than distributed through it.
Quantify the affected proportion, which determines whether the timing analysis is usable at all.
Then find out why, which is usually a device, a process step or an ergonomic problem rather than a discipline one.
Making the scan useful to the scanner
Scan discipline follows from whether the scan does something for the person performing it.
Confirmation they need: the right item, the right quantity, the next location.
Immediate feedback, not a silent accept.
A short interaction. Every extra second is multiplied by thousands of scans a shift.
A device that works one-handed, because the other hand is holding something.
Where the scan exists only to feed a report, it will be batched, and no amount of instruction changes that.