Nobody builds a distribution centre network on an empty floor. By the time someone calls us, there is already a warehouse running, orders going out, scanners in people's hands, and a network that has been extended one access point at a time over several years by whoever was available.
The request is never "build us a network." It is "fix this, and do not stop us picking." Those two things are in tension, and how you handle that tension is most of the job.
Here is how we approach it, and where it tends to go wrong.
Start by finding out what is actually there
The documentation, if it exists, will be wrong. This is not a criticism of anyone. Networks that grow by accretion always end up differing from whatever was written down, because the changes that get made under pressure are the ones that never get recorded.
So the first pass is discovery, on foot and on the wire. What switches exist and where. What is plugged into what. Which access points are live, which are powered but dead, and which have been unreachable so long that nobody remembers they are there. What address ranges are in use. Where the wireless coverage genuinely reaches with racking full, which is a different map from the one you get with racking empty.
That last point matters more in a warehouse than anywhere else. Stock absorbs and reflects signal. A coverage survey done during a quiet period, with half-empty racks, will tell you the site is fine. Then the containers land, the aisles fill, and the dead zone appears in the middle of the pick face.
Walk it loaded. If you cannot walk it loaded, walk it twice and compare.
Build alongside, do not rip out
The instinct is to replace the old network with the new one. On a live site, that is how you end up explaining why picking stopped.
The approach that works is to build the new network in parallel with the old one and move onto it in stages. New switching goes in beside the existing gear rather than instead of it. New access points go up while the old ones stay live. For a period, both networks exist at once. That costs a little more in hardware and temporary cabling, and it buys you the ability to stop at any point without having broken anything.
It also changes the psychology of the work. If the new network is being built alongside a working one, a problem discovered at 2pm is a problem you fix tomorrow. If you have already removed the old one, it is a crisis.
Cut over in zones, at the quietest hour, one at a time
Once the parallel network is up and tested, move onto it zone by zone: receiving, then one pick area, then the next, then despatch and the offices. Never the whole floor at once.
Each zone cutover follows the same pattern: move it, watch it through a full working period, and only then schedule the next one. Watching means actually watching: someone on the floor with the people using it, not a dashboard check from an office.
Pick the window honestly. Every warehouse has a genuine quiet hour and it is rarely the one on the schedule. Ask the supervisors, not the plan.
What to test before you move anything
Testing a warehouse network by browsing a website on a laptop tells you almost nothing. The things that break are specific:
- Scanners, on the move. A handheld that works standing still under an access point is not a test. Walk the full route a picker walks, at picking pace, through the whole zone, and watch what happens as the device moves between access points. Roaming is where warehouse wireless actually fails.
- Every device class, not one of each. The old scanners and the new ones behave differently. So do the forklift-mounted terminals, the label printers, the tablets in receiving, and whatever one-off device the returns desk uses.
- Printers. Label printers are the most reliably annoying device in any warehouse cutover. They hold static addresses set years ago by someone who has left, they are shared by several systems, and they fail silently. Confirm every one of them before you touch the zone it lives in.
- The systems, not just the connection. The warehouse management system, the scanning application, and anything that talks to head office or a courier. A device can be perfectly on the network and still unable to reach the one system that matters.
- What happens when the internet drops. Find out deliberately, in a controlled test, which functions keep working on the local network and which stop. Better to learn this at 6am on a Tuesday than during a peak week.
Have a way back
Every zone cutover needs a rollback that can be executed by whoever is on site, in minutes, without a phone call. Usually this is simple. The old access points are still up, the old switch port is still live, and moving back is a matter of reverting one change. But it needs to be decided and written down before the cutover, not improvised.
Set the trigger in advance too. Agree what "this is not working" looks like before you start, because in the moment there is enormous pressure to keep troubleshooting a live problem rather than backing out and trying again tomorrow. A rollback is not a failure. Stopping despatch for three hours is.
The three that bite
Address ranges. Old networks accumulate address ranges nobody documented, often with devices holding fixed addresses inside a range that is also handing out addresses automatically. Plan the new addressing properly and check the old ranges for fixed devices hiding in them before you reuse anything.
Roaming. Devices that hold onto a distant access point rather than switching to the strong one overhead cause the classic symptom: the scanner works everywhere except one aisle, intermittently. This is an access point placement and configuration problem, and it is the single most common reason a warehouse network is disliked by the people using it.
Devices nobody owns. The camera recorder, the time clock, the alarm panel, the compressor monitor. Each one is on the network, each one belongs to a different supplier, and none of them appear on any list. Find them during discovery. They will not tolerate being surprised.
When it is done, leave it maintainable
The last mile is the part that gets dropped when everyone is tired and the site is working again:
- Label every port, every cable, and every access point, physically, at both ends.
- Write down what is where and what it does, on one page, in a place the site can actually find it.
- Put monitoring on the things whose failure is currently only detected by someone complaining.
- Keep spares on site for the parts that stop the operation, and make sure someone there knows how to swap them.
A warehouse network that only one person understands is a network with a single point of failure that does not appear on any diagram.
The honest summary
Doing this without downtime is not a technical trick. It is a sequencing decision: build alongside, move in small pieces, test the way the equipment is actually used, and keep a way back at every step. It takes longer than a rip-and-replace and it costs slightly more in temporary hardware. It also means the operation never stops, which is worth considerably more than the difference.
If you are looking at a warehouse or depot network that has grown past what it can handle, this is the work we do, including the part where nobody notices it happened.