Website monitoring: knowing your site is down before your customers do

An uptime promise is about averages; your site is either up for a customer right now or it is not, and only something checking from outside knows which.

The worst way to learn that a website is down is from a customer, because the customer learned it before you did, and they learned it at the moment they needed the site to work. Every owner knows this. Very few own a check that would have told them first, and the gap between those two facts is where silent downtime lives.

The honest framing separates two different promises. The host's uptime promise is a figure about averages across a month, with definitions attached and compensation that rarely arrives on its own. Your reality has no averages: at any given minute, the page a customer loads either answers or it does not. The only way to know which is to have something outside your own building request the page the way a visitor does, and tell you when it stops.

The failure versions repeat because each one feels like diligence at the time. The site that went down on Friday evening and stayed down through the weekend, discovered by Monday's angry message after the inquiries were already lost. The owner who checks the site manually every morning, and stops checking within a fortnight, because the quiet days erode the habit one uneventful morning at a time. And the monitoring that watches the server's heartbeat while the page a customer actually loads fails silently beside it.

The practice has three parts, and none of them is exotic. An external check on the real URL, the address customers actually type, not an internal dashboard nobody else sees. An alert that reaches a phone somebody actually carries, because a report nobody reads is downtime with paperwork attached. And a second check on the page that costs money when silent: the contact form or the checkout, exercised the way a visitor exercises it, so that a page which loads but cannot accept a message counts as down, which is exactly what it is.

Then the alerts themselves have to stay trustworthy. A check that cries wolf at every network hiccup gets muted within the week, and the muting is worse than the noise, because now nothing is watched at all. Thresholds get set so that a real failure pages you and a passing blip does not; and when an alert does fire, someone acts on it, or the whole apparatus becomes a night light. Monitoring is a habit with machinery attached, not machinery alone.

We watch our own properties this way, from outside, on the pages that matter, which is how we know the alerts are worth the phone they ring. When we take over a client's site, wiring the external check is part of the handover, because a site we built and left unwatched would be a failure kept private. Guidance, not guarantees: monitoring tells you that something broke, and tells you fast; it does not fix the thing, and it does not replace the copies you will need when the break is real.