Key takeaways
- Server uptime does not tell you the checkout works
- Check the homepage, a product page, add to basket and the checkout
- If everything is urgent, nothing is, and people stop reading the channel
- Name who receives alerts, what they do, and who they escalate to
- Trigger an alert deliberately to prove the path works out of hours
- Carry a correlation identifier, never customer names or order contents
Most ecommerce sites have monitoring. Rather fewer have monitoring that reaches a human being at the moment it matters.
The difference is not tooling. It is deciding, in advance, which conditions justify interrupting someone.
Monitor the transaction, not just the server
Server uptime tells you the machine is on. It does not tell you that the checkout is broken.
The checks that matter are the ones that follow the money:
- The homepage returns 200
- A product page returns 200 and shows a price
- An item can be added to a basket
- The checkout loads and the payment provider responds
The fourth is the one people skip, because it needs a controlled test mode. It is also the one that catches the failure that costs the most.
Separate the urgent from the interesting
Not everything deserves to wake someone at three in the morning.
| Level | Examples | Response |
|---|---|---|
| Wake someone | Checkout failing, site down, payment provider errors | Immediate |
| Notify in hours | Elevated 404s, slow category pages, failed stock sync | Same working day |
| Report weekly | Crawl anomalies, dependency updates, performance drift | Reviewed, not reacted to |
If everything is urgent, nothing is, and people start ignoring the channel. That is how an outage lasts nine hours.
Name the person
An alert needs a destination that is a person, not a shared inbox nobody owns. Write down:
- Who receives alerts, by level
- What they are expected to do
- Who they escalate to
- What happens when that person is on holiday
This is a five-minute conversation that gets skipped for years.
Test the alert path deliberately
The only way to know an alert works is to trigger one on purpose. Break something in a controlled way, confirm the message arrives, and confirm it arrives somewhere a human will see it outside working hours.
Do this when you set it up, and again whenever the team or the tooling changes.
Record the correlation, not the customer
Alerts and error logs should carry a correlation identifier that lets an engineer find the request. They should not carry names, email addresses or order contents. You want to be able to debug an incident without creating a second copy of your customers' personal data in a monitoring tool.
Frequently asked
Questions readers ask
What should I monitor on an ecommerce site?
Follow the money: the homepage returns 200, a product page shows a price, an item can be added to a basket, and the checkout loads with the payment provider responding. The last is the one people skip and the one that catches the costliest failures.
What conditions justify an out-of-hours alert?
Checkout failing, the site down, or payment provider errors. Elevated 404s, slow category pages and failed stock syncs belong in a same-day notification. Crawl anomalies and performance drift belong in a weekly report.
How do I know my alerts actually work?
Trigger one on purpose. Break something in a controlled way, confirm the message arrives, and confirm it arrives somewhere a human will see it outside working hours. Repeat whenever the team or tooling changes.





