The Data You Already Have

The instinct is to install something. The higher-return first move is joining what the operation already generates.

The sources

Warehouse management system events. Task created, assigned, started, completed, with timestamps and locations. The richest source in most facilities and routinely used only for operational control rather than analysis.

Scan events. Every barcode and RFID read, with time, location and item. These bracket every handling step.

Equipment telematics, on modern handling equipment: hours, movement, impacts, battery state.

Labour management data, where a system exists.

Access control and gate records, which give arrival and departure at the facility and yard level.

Order and despatch records, which bracket the whole flow and give the demand curve.

Maintenance records, which explain equipment gaps.

Joining them

Time is the primary key, and it requires the clocks to agree. Check this first; unsynchronised clocks across systems make every cross-system analysis wrong in ways that look plausible.

Location is the second, and it requires a consistent location scheme. Most facilities have several, historically.

Item and order identifiers link the flow end to end.

Equipment identifiers link telematics to tasks.

The join is the work, and it is mostly reconciling identifiers that were never designed to be joined.

The first analyses

Achievable within days once the join exists.

Elapsed time decomposition for a unit of work.

Queue length by stage over the day, which localises the bottleneck.

Demand curve by hour, day and season, against staffing.

Travel implied by task sequence, from location data.

Exception and rework rate by product line and by stage.

Equipment utilisation by hour.

None of these requires new hardware, and together they usually redirect the project's priorities.

What existing data cannot tell you

Worth establishing before assuming a new sensor is needed.

What happens between scans. The gap between two reads is a black box: travel, waiting, searching, or something else entirely.

Where people and equipment actually are in real time, rather than at discrete read points.

Why an exception occurred.

Near-misses, which by definition leave no transaction record.

Manual handling and posture, which no transactional system sees.

These gaps are the legitimate case for new sensing, and identifying them specifically produces a much better-scoped project than starting from a product.

The data quality reckoning

Every facility discovers the same problems at this stage.

Scans missed or performed in batches, so timestamps do not reflect reality.

Location master data out of date.

Tasks left open, inflating elapsed times.

Clock drift between systems.

Manual corrections that overwrite the original event.

Fix these before drawing conclusions, because the analysis will otherwise produce confident findings about artefacts. This has its own note.

The sequencing point

Join, analyse, identify the specific gaps, then buy for those gaps.

Projects that start with a purchase acquire a system scoped to what the vendor sells rather than to what the facility cannot currently see.

Checking clock agreement first

The single check that prevents a whole class of confidently wrong results.

List every system contributing a timestamp.

Query the current time from each, simultaneously.

Record the offsets.

Anything beyond a few seconds makes cross-system timing analysis wrong in ways that look entirely plausible.

Fix by synchronising to a common source, which is a configuration change.

Re-check quarterly, because drift returns and because new systems arrive unsynchronised.

Do this before the first analysis, not after someone questions a result.

Building the location scheme

Facilities accumulate several location naming schemes historically, and every cross-system join depends on reconciling them.

List every scheme in use: the management system, the racking labels, the floor markings, the fire plan.

Map them to one another, which is a walk as well as a spreadsheet.

Add coordinates, even approximate ones, which is what makes distance calculation possible.

Record aisle, bay, level and side as separate fields rather than as one string.

Maintain it, because racking changes and a stale location scheme silently corrupts travel analysis.