Ten orders didn’t make it to the ERP, and it was days before anyone noticed. No alert went off at any point - someone spotted the gap while reconciling by hand and pulled the thread from there.
When they dug into it, the integration hadn’t thrown a single error, because it had never picked those orders up in the first place.
It’s an easy situation to end up in before peak. The connectors are running and there’s nothing red in the logs, so the reasonable assumption is that orders are moving.
This one’s for whoever owns the chain from storefront to ERP over the next few weeks. That’s usually someone in operations or finance, being told the integrations are fine, who’ll have to explain the numbers in December if they weren’t. With Black Friday on the 27th, there’s still time to check whether your monitoring would catch a missing order, and whether anyone could get it back through the chain cleanly.
When nothing fails
Most setups have some exception alerting, so when an API call fails or a payload doesn’t validate, an email or a ticket goes to someone. It won’t tell you about the flow that timed out overnight and never restarted, or the webhook that stopped firing after a platform update, because in both cases nothing was attempted and so nothing failed.
It’s worth asking your integration partner, or whoever runs this internally, how they’d find out about an order that never arrived. If the honest answer is that someone would eventually notice, that’s the gap to close before peak.
Who picks up the alert
A lot of alert routing was set up once, during the original build, and never looked at again. Everything goes to a shared mailbox that customer service keeps half an eye on, and it’s assumed someone will escalate if it looks serious.
Before peak, take one example alert and follow it through. Someone needs to see it and work out the cause. If the source data is wrong, someone has to fix it, and a replay needs someone with the authority to approve it. After that, somebody has to check the destination is right and that nothing posted twice.
In a small team these are often the same person, which is fine as long as everyone knows it’s them.
If a stuck order flow sits in the same Slack channel as a hundred low-stock warnings, people stop reading the channel fairly quickly. An urgent alert should say which flow and which orders are affected, and how long it’s been going on. Ideally it should name who owns the next step as well.
Practising the replay
The replay process is the part we’d most want tested before peak, mainly because a retry under pressure can double-post revenue or duplicate refunds, and unpicking that can take finance a week.
So run through a few failures somewhere safe while there’s still time, using a sandbox where you have one and keeping anything in production small. Good ones to try are:
A flow that stops completely
A transaction that fails validation
A slow destination with a queue building behind it
A required state arriving late
A simple replay of a failed order
Walk each one from the alert through to reconciliation, and only sign it off once the destination is correct.
Peak is mostly an integration test
Most of the pre-peak attention goes on the storefront, which makes sense since that’s what customers see. Integration work can be tedious and hard to show, but when done properly, you hardly need to open the integration platform in November because the controls tell you what’s going on and the team has rehearsed what to do.
If you’re the one who owns that chain, the next fortnight is the last sensible window to test the monitoring and the recovery as well as the build, and to find out whether ten missing orders would reach you the same morning or five days later, from someone with a spreadsheet.



