Let me give you the conclusion first. When an integration goes down, the first person to know should not be the customer or the warehouse floor. It should be your monitoring system. If your team only finds out because a customer complains, your monitoring is decoration. I say this before every go-live, and I repeat it on go-live night, because I have seen this failure three times. The cost was different each time, but the root cause was the same every time: nobody treated alerting as a serious piece of work.
The first time was at an e-commerce warehouse. The order interface between the WMS and the marketplace went down at two in the morning. Nobody knew. At eight the warehouse opened, pickers noticed the day's orders were far fewer than usual, and someone joked they would get off early. It was not until ten o'clock, when customer service started getting calls asking why orders had not shipped for two days, that anyone realized the interface had been down for eight hours. By then the backlog had passed a thousand orders and the support phones were overwhelmed. That was the first time I saw this failure.
The second time was more typical, and harder to catch. The interface did not fail completely; it dropped orders intermittently. The dashboard stayed green, but every day a few percent of orders quietly disappeared. Customers complained sporadically, the warehouse assumed they had failed to place their orders, and it took almost two weeks to discover the interface was timing out at peak hours, with some orders never writing back. An intermittent failure is harder to find than a total outage, because it never trips any single alert threshold.
The third time was the opposite problem: too many alerts. A phone buzzed dozens of times a day, almost all of it noise. The operations engineer got fed up and muted the alert group, so on the day something genuinely broke, the alert went out and nobody saw it. Monitoring that has been muted is the same as no monitoring at all.
Three failures, three ways to die, but one lesson: how you set up WMS monitoring and alerting decides who finds out first when an integration goes down.
What WMS Integration Monitoring Should Watch: Four Signals, and You Need All of Them
Setting up WMS integration monitoring and alerting starts with deciding what to watch. On my projects, I ask clients to watch at least four signals.
The first is connectivity. Can the interface be reached at all? Probe the partner interface on a schedule: a successful probe gets logged, a failed probe alerts immediately. And the probe must check business connectivity, not just network connectivity. You need a real response back before you can call it healthy. Many projects probe only the network, and the partner service returns an error code while the network stays up, so the monitoring misses it entirely.
The second is success rate. Order push, inventory sync, shipment callback, each interface needs its own success rate. When the rate drops, even by a few percentage points, it should alert. The intermittent order loss I described earlier was a success rate quietly sliding with nobody watching a threshold. My approach is to set thresholds per interface, stricter for core interfaces, looser for supporting ones.
The third is latency. The interface is up, but response time has gone from one second to ten seconds. That is a warning sign, and at peak hours it matters most: once latency climbs, timeout losses follow immediately. Latency thresholds need to differ between peak and off-peak. One threshold for the whole day false-alarms in quiet hours and alerts too late at peak.
The fourth is backlog. How many messages are sitting in the queue, unprocessed? Under normal conditions the backlog should fluctuate within a few dozen orders. Once it passes a few hundred and keeps rising, the downstream side cannot keep up. The advantage of this signal is that it does not depend on the partner's response: even if the partner service is completely silent, the backlog still grows on its own.
You need all four signals together. Leave one out and you have a blind spot: connectivity tells you whether it is up, success rate whether anything is being lost, latency whether it is fast enough, backlog whether it is jammed. Only the four together give you a full picture of integration health.
How to Set Up WMS Integration Monitoring and Alerts: Severity, Owner, and Time of Day
Once you know what to watch, the next step is how to configure the alerts. My rule is three phrases: severity, owner, and time of day.
Severity means splitting alerts into three levels. Level one is a core interface down, a success rate collapse, or a backlog climbing without stopping. That goes out by phone and text to the on-call lead immediately, even in the middle of the night. Level two is latency over threshold or a small success-rate dip, sent to the work group for the on-call engineer to handle during working hours. Level three is informational, for example a single partner timeout that succeeded on automatic retry. That goes into the log only and should not disturb anyone.
Owner means different levels go to different people. A level-one alert must reach someone who can make decisions, not just the operations engineer, but the operations lead as well. An interface outage affects shipping, so the operations lead has to know right away and decide whether to start the manual fallback plan. Level-two alerts can go to operations and implementation engineers. Level-three alerts go to nobody; they are there for the review meeting.
Time of day means nights are handled differently from days: level one gets an immediate response day or night, level two can wait until morning. Wake someone at three in the morning for an informational message twice and they will mute the group, and once muted, nobody sees the real alert.
The Go-Live Night Checklist: Go Through It Item by Item
Every go-live night, I sit with the client and walk through the list below. I have seen this failure three times, so every item on it was paid for with real money.
First, is the health probe actually running? Break the interface manually and check whether the alert arrives within a few minutes. An alert that has never been tested is no alert at all.
Second, are success-rate thresholds set per interface? If core and supporting interfaces share one threshold, split them.
Third, are latency thresholds split between peak and off-peak? Compare them against last week's peak latency data.
Fourth, what is the backlog alert line? How many orders, and for how long, before it alerts? Write that number down next to the on-call rota.
Fifth, are the alert recipients correct? Can the level-one phone number be reached? Has the on-call rota changed, and was the notification list updated? An expired contact list is one of the most common failure points.
Sixth, is there a quiet period and deduplication? The same fault should not alert ten times in five minutes. Repeated alerting is the fastest way to get the channel muted.
Seventh, is the fallback plan written down? If the interface is down, how does the warehouse ship manually? If inventory is out of sync, how does operations reconcile it? The end point of an alert is not "we know." It is "someone acts on a plan."
Only after those seven items pass will I let a project go live.
Three Failure Points People Most Often Miss
Let me close with three failure points people most often miss. The first is monitoring only your own system, not the partner interface. Everything can look fine on the WMS side while the partner is down, and monitoring has to cover both sides, at least as far as the partner's business response.
The second is alerts with no owner. An alert goes to a group, everyone can see it, and nobody handles it. Every alert level needs a named owner, with a name and a phone number on the rota. A message in a group is not the same as someone being responsible.
The third is never reviewing alerts after go-live. Alert logs need regular review: thresholds that false-alarm too often need adjusting, scenarios that slipped through need adding. Monitoring is not a one-time setup. It has to grow with volume, with the number of interfaces, and with changes in partners.
When an integration goes down, who finds out first? The answer decides whether your warehouse fights the fire proactively or takes the blame passively. With monitoring set up properly, you know within three minutes and fix it before the customer notices. Without it, you find out eight hours later, when the damage is done.
On go-live night, walk through this list. One hour can save you an eight-hour outage and a thousand-order backlog.
By Lin Xiao, WMS Implementation Consultant







