Gartner published a piece on 16 September on why AI strategies fail even when the budget is there. Two figures in it matter for anyone planning warehouse software next year.

Seventy-six percent of CEOs in the 2026 Gartner CEO and Senior Business Executive Survey name AI as the technology most likely to disrupt their industry over the next three years.

And, in Gartner's words, “59% of AI pilots fail to reach production, primarily because of organizational, not technical constraints.”

David Furlonger, Distinguished Vice President Analyst and Gartner Fellow, described the risk like this:

“The real danger for an enterprise in today's AI consensus is not an underinvestment in AI but an overconfidence that technology can compensate for leadership, infrastructure, and operational and technology systems that were never designed for algorithmic speed.”

Gartner is writing about enterprises in general. Warehouse AI follows the same pattern, and the starting point is easy to find.

What slotting and cartonisation software reads

Slotting, pick routing and cartonisation tools make their decisions from the item master.

To slot a SKU, the software needs the dimensions and weight of every unit of measure the item is handled in, the pack hierarchy from each to inner to case to pallet, and velocity from order history. To choose a carton, it needs the same dimensions plus your carton range.

When those fields are wrong, the software makes wrong decisions quickly and presents them as recommendations. Gartner's phrase for this is “faster, less transparent errors”.

Three ways the item master fails

Incompleteness. Fields are blank. A filter finds them in minutes.

Inaccuracy. Fields are filled in and wrong. A case dimension smaller than the units inside it. A case weight copied from the each. A pallet height that leaves out the pallet. Every one of these passes a completeness check and fails on the floor.

Duplication. The same product sits under two codes, velocity is split across both, and neither looks fast enough to earn a pick-face slot near the dispatch end.

Incomplete data and wrong data look the same on a screen: a full grid.

That's why a pilot can pass its demo on a sample file and stall once it runs against the full item master.

Where pilots stall in a DC

Gartner's main point is organisational. In a warehouse, that turns into three questions that often go unanswered until the pilot is already running.

Who owns the item master? Procurement creates the item, the DC measures it and IT holds the system. In many operations nobody signs off the dimensions.

Who approves a re-slot? The software recommends moving 400 SKUs. The DC manager, the inventory controller and the night-shift supervisor all have a say, and if nobody owns the decision the moves don't happen.

Who measures the result? If walking distance per order line wasn't recorded before the pilot, there's no baseline to prove a gain against.

Gartner says leaders who “clarify decision rights will have the advantage.” In a DC, that means a named owner for item data and a named approver for slot moves, both written down before go-live.

Keep the WMS, keep the data portable

Gartner also recommends avoiding commitment to one vendor by “using flexible APIs, abstraction layers and portable data architectures.”

For South African DCs, that supports adding an optimisation layer to the WMS you already run, whether that's Körber, SAP EWM, Manhattan or another platform, and keeping the WMS as the system of record. The item master stays in one place. The layer reads it and sends recommendations back.

Optioryx Pulse works this way. Optioryx states that Pulse is not a WMS and does not replace one. It connects through manual file import and export, flat files over CSV or SFTP, or a direct API into your WMS, TMS or ERP. The item data you clean for it stays usable by whatever system you run next.

Three things to do before you fund a pilot

If your financial year starts in April, FY2027/28 budgets are being drafted now. Do these three before an AI line goes into that budget.

1. Check the item master for the SKUs that carry most of your picks. Start with completeness, then run sanity checks: each case larger than the units inside it, weight rising at every level of the pack hierarchy, pallet height that includes the pallet. Our free Item Tracker scores completeness per item. The sanity checks are a separate pass.

2. Name two owners. One person signs off item dimensions. One person approves slot moves.

3. Record a baseline. Pick 2 or 3 measures, such as walking distance per order line, lines picked per hour and cartons per order, and record them for 4 weeks before anything changes.

A pilot that starts with those three in place has a much better chance of reaching production.

Sources