Bottom line first: a demo only proves a WMS looks good. A WMS POC is what proves it actually runs in your warehouse. Sign after a demo and nothing else, and you are buying blind. When go-live blows up, it will not be bad luck — it will be the missing step: two weeks with your real data, your real exception orders, and your floor staff running the system.

Supply & Demand Chain Executive, citing Georgia Tech research, reports that more than 30% of WMS implementations fail, only 42% finish within budget, and 56% run late. A typical project takes 9 to 12 months to implement and 24 to 36 months to pay for itself; about 30% miss go-live, then need another 3 to 12 months to become fully operational. Pick wrong and you pay for one to three years.

Why demos look so good: you never see the script

A demo looks good because the stage is set. Clean master data runs a clean process. The picker taps twice, goods move from point A to point B, a green checkmark pops up. Everyone nods.

No real warehouse runs a standard process. A real one has this: a customer changes the delivery address mid-wave, an urgent order drops in halfway through picking, stock with 20 days of shelf life sits mixed in with stock that has 200, and the system says five on the shelf while the shelf holds four. A demo never shows you any of that. Exception orders are hidden from the script entirely. I have seen this exact failure three times — applause on demo day, chaos in week one of go-live. The salesperson is not a villain; he is showing you the system's best side. If that is the only side you look at before signing, the failure is your choice.

Two go-lives that went sideways

In October 2023, a 3PL handling auto parts in Kunshan — 12,000 square meters, about 4,000 orders a day — picked the smoothest of three demos. I asked one question: how does it handle expiry dates and short picks? The salesperson clicked twice, an exception screen appeared, and it looked convincing. The client signed on the spot.

Week one broke the operation. The system pushed locations first-in-first-out and kept pushing stock with one month of shelf life left, while customers needed six months or more. Pickers overrode manually, the system refused, and verification jammed. Out of 4,000 daily orders, more than 600 got stuck at the verification bench. The warehouse manager called me at midnight: nothing was shipping. We spent three weeks rebuilding expiry rules, redoing verification logic, and recounting over 2,000 locations. The go-live-night checklist had a line for "expiry rules tested with real data." Nobody had run it during selection.

In March 2024, a beauty e-commerce warehouse in Guangzhou — 6,000 orders a day, 12,000 in peak — ran a POC, but the wrong people ran it. Only two IT staff touched the system; no picker, verifier, or putaway operator got near it. IT ran standard orders, and the report looked beautiful. Acceptance passed, go-live. On day one, pickers stood at the racks staring: screen location codes did not match shelf labels, wave tasks dropped 80 orders at once, and pickers pushed carts back and forth down the aisles. Picking output per person fell about 40%. The system was not broken — it had never been designed around how floor staff actually move. We split the waves smaller, relabeled the locations, and took two more weeks to pull throughput back.

Same story twice: great demos, a POC either skipped or faked.

How to run a WMS POC: five steps across two weeks

A WMS POC is not a second demo. It drags the system into your real environment and forces it to show its true colors. Two weeks is enough — but only if you run them this way.

Step one: bring your own data, never the vendor's sample database. Export your real SKU list — at least 2,000 SKUs — and three months of real historical orders — at least 5,000. Then prepare 200 exception orders: mis-ships, short picks, near-expiry stock, address changes, rush orders inserted mid-wave, returns put back to stock — 20 of each type, minimum. If a vendor says the sample data is enough, find another vendor.

Step two: write acceptance criteria on paper, with numbers. I use these five as a starting point: picking accuracy 99.5% or better; putaway within 4 hours of receiving; exception orders closed within 2 hours; a rush order re-planned within 15 minutes; stocktaking variance under 0.3%. Adjust them to your business, but write them as hard figures. "The system should deliver high accuracy" is the same as writing nothing.

Step three: lock the two-week time box. Week one runs the normal flow end to end: receiving, putaway, picking, verification, packing, shipping. Week two runs exceptions only: all 200 exception orders, plus one simulated peak day with volume doubled. At the end of week two, pass means move forward, fail means stop. There is no "one more week."

Step four: put floor staff on the system, not just IT. Pickers, verifiers, putaway operators, and the warehouse supervisor — each runs at least two full days of real tasks. IT checks interfaces and data; operations checks usability. The questions floor staff ask after two days are worth ten times an IT test report. When a picker says "I cannot read this screen," the system needs to change — not the picker.

Step five: write the exit clause into the POC contract — fail it, walk away. Three lines: the acceptance criteria are the five numbers from step two; passing means three consecutive days of measured data, not a vendor promise; if it fails, the client may terminate — the POC fee stays paid, but no implementation fees follow. The clearer the clause, the less a vendor dares to fake it.

Do the math: what a WMS POC is really worth

The table uses a mid-size warehouse at 3,000 orders a day. The figures are a worked example — plug in your own volume and rerun it. POC cost is pocket change next to a failed go-live.

Item Sign after demo only Run a two-week POC first
Selection-stage spend 3 days of demos, near zero Two-week POC, roughly $11,000–$17,000
Mis-ship rate after go-live Around 2%, three months of rework Under 0.5%, stable in two weeks
Rework cost ($2 per reworked order) About $3,600 a month About $900 a month
Delayed go-live 1–3 months late; hit peak season and it is a disaster Mostly on schedule
Cost of switching systems Contract signed; switching means redoing the project Fail the POC, walk away; the loss is the POC

Here is the counterintuitive part: a POC that goes perfectly smoothly should worry you. Two weeks, zero exceptions — your exception orders were not tough enough, or the vendor was performing for you. A real POC should surface problems. That is the entire point.

The go-live-night checklist: tick every box before you sign. This is the checklist I run the night before every go-live. POC acceptance goes through the same list — one box unticked, do not sign.

  • Full process run on real data; all 200 exceptions closed
  • Five acceptance criteria met on three consecutive days
  • Every floor operator ran two days of real tasks; feedback recorded
  • Expiry, batch, and serial rules tested against real rules
  • Rush orders, address changes, and returns all tested live
  • ERP and parcel-carrier interfaces tested with real orders
  • One full stocktake cycle run, variance within 0.3%
  • Exit clause: fail the criteria, walk away

Five failure points: if any one shows up, do not sign.

One: the vendor says "that feature comes in phase two." Phase two means unscheduled. Core features must pass in the POC.

Two: the POC runs the happy path only. Exceptions are the daily reality of a warehouse — a POC without them tests nothing.

Three: acceptance based on demo-day performance. One day of data proves nothing — judge three consecutive days.

Four: vendor staff run the POC for you. A vendor consultant knows the system ten times better than you do. Smooth in his hands says nothing about your staff's. Your own people must run it.

Five: no exit clause. Without one, even a rigorous POC is useless — fail the acceptance and you still go live with it.

A POC is your test ground: two weeks and a few thousand dollars buys one answer: does this system actually work here? It is the cheapest money in the whole WMS project. Pull your real order data tomorrow, prepare 200 exception orders, and only then book the POC.

By Lin Xiao, WMS Implementation Consultant