Cubiscan publishes a throughput figure for its item master dimensioner: one person measures 450 to 500 SKUs in an 8-hour shift.

Take the midpoint. A distributor carrying 8,000 active SKUs needs about 17 operator shifts to rebuild the file. That is three and a half weeks of one person, and it fits comfortably inside the 14 weeks to 27 November.

A distributor carrying 30,000 SKUs needs about 63 shifts. One person, five days a week, is 13 weeks. The entire remaining runway, with nothing left for validation, load, or the exceptions that always turn up.

So at the larger SKU count the honest answer is that the whole file does not get rebuilt before peak. What gets decided instead is which SKUs get measured, in what order, and what happens to the rest.

What a wrong dimension does downstream

A WMS is a rules engine. It does not pause when an item attribute is wrong. It keeps executing, faster and more consistently than a person would, on the number it was given.

Cubiscan sets this out plainly in its own material: bad data causes the system to automate incorrect decisions faster, which increases exceptions and rework. Four places it shows up.

Slotting puts a SKU in a location it does not physically fit, which generates relocations, aisle congestion where oversized items land in a pick module, and poor cube use in storage.

Replenishment fires at the wrong quantity when case pack or conversion factors are inconsistent. Forward pick locations run empty while bulk holds stock.

Labour planning drifts, because the system cannot estimate work content for an item whose size and weight it has wrong. Wave planning then compounds it when carton fit assumptions fail.

Pack-out throws exceptions. The carton will not close, the void fill goes up, and the carrier reweighs and bills on corrected dimensions rather than the ones you quoted.

Each of those is survivable in September at normal volume. In the last week of November, at 3 to 4 times normal outbound, each one is a queue.

Why the file looks fine

Item master data fails three ways: it is incomplete, it is inaccurate, or it is duplicated.

The difficult one is the second, because incomplete data and wrong data look identical on a screen. A blank length field is obvious to anyone who opens the record. A case dimension that is smaller than the units inside it reads as complete, passes every "field populated" check, and fails on the floor.

The unit of measure hierarchy is where this concentrates. Each, inner, case, pallet. Dimensions and weight at every level, with conversions that reconcile. If the each is right and the case is an estimate someone typed in 2023, cartonization has nothing to work from and slotting inherits the same error.

Two checks separate the two failure modes without measuring anything.

Records that fail either test are wrong, not missing, and they are the ones costing money today. That query runs against an export in an afternoon.

The 14-week version

The scoped plan, assuming one operator and a mixed SKU base.

Weeks 1 and 2. Rank and reconcile. Pull 12 months of outbound order lines. Rank SKUs by pick frequency, not by revenue, because the file is being fixed for physical work rather than for reporting. Run the two reconciliation checks above across the whole file. You now have two lists: the SKUs that move, and the SKUs whose data is provably wrong.

Weeks 3 to 8. Measure the intersection first. SKUs that are both high frequency and provably wrong come first. In most operations that is a few hundred records and it takes a fortnight. Then work down the frequency ranking. In a typical distribution profile, 20% of SKUs carry 70 to 80% of pick lines, so 6 weeks of measuring against a frequency ranking covers most of what peak will actually touch.

Weeks 9 and 10. Load and validate. Write the corrected records back, then sample. Pull 50 SKUs at random from the corrected set, measure them again independently, and compare. If the resample disagrees on more than 2 or 3 of the 50, the measuring method is inconsistent and the rest of the file inherits that.

Weeks 11 to 14. Freeze and govern. No bulk changes inside the last month before peak. New SKUs get held for measurement at receiving before they are released to active storage, which stops the file degrading while you are watching it. Pack station exceptions get logged by SKU, and repeated carton-fit failures trigger a remeasure.

The freeze matters. A dim file change in the third week of November alters carton selection and slotting recommendations at the worst possible moment.

Where this connects to the software layer

Cartonization, slotting and load building all run off the same fields. Optioryx publishes 15% lower shipping cost, 30% less carton and 35% less air shipped for its Pulse 3D Cartonization module, and 20% higher picking productivity with 40% shorter walking distances for Pulse Pick. Those modules connect on top of your existing WMS through an API. Optioryx states that Pulse is not a WMS and does not replace one.

Read those percentages as direction rather than a promise for your building. What matters more here is the dependency. Every one of those numbers is computed from item master dimensions and the UoM hierarchy. Clean the file and the software has something to work with. Leave it and the optimisation layer applies precise logic to wrong inputs, which is worse than the manual process it replaced, because it is faster and nobody is checking it.

That is the sequencing argument for anyone considering a cartonization or slotting project in 2027. The measurement work is the part with a lead time, and it is the part you can start this week without a budget approval or a vendor.

The sustaining mechanism is separate and comes later. Mobile dimensioning at inbound, capturing accurate length, width, height and weight on a device the receiving team already carries, writing clean records back to the WMS. That keeps the file from decaying again after the rebuild. It is the second-year conversation, not the one for the next 14 weeks.

The commercial context

South African online retail is heading past R150 billion in 2026 and now sits near 10% of total retail sales. That volume does not arrive evenly. It concentrates into the window from late November through the January returns wave, which is the same window in which a wrong case dimension turns into a queue at pack-out.

Fourteen weeks is enough time to fix the SKUs that carry the volume. It is not enough time to fix all of them, and pretending otherwise is how operations arrive at the third week of November with a half-measured file and no ranking.

Rank by pick frequency, reconcile the hierarchy, measure the intersection, then freeze.

Get the reconciliation query and the ranking method

Both run off an item master export and 12 months of order lines you already have.

Book a master data check Explore cartonization

Sources

Note on the throughput figure: Cubiscan publishes 450 to 500 items per 8-hour shift for one operator on a Cubiscan 325. The 475 used throughout this article is the midpoint. Your own rate will depend on item mix and handling.