How often your site goes down, and what that actually means

There is a category of article you recognise instantly: the ones that open with “every minute of downtime costs you thousands”. Let me disappoint you early. You will not find a single such figure in here, and I will explain why at the end.

That does not mean the subject lacks substance. It has plenty, and measurable at that. The substance simply is not in dollars nobody has seen, it is in three things you can see: what your provider actually promises, what your site does when it falls over, and how you find out.

First the arithmetic, which needs no source

“99.9%” is one of the most successful commercial tricks I know, because it looks like 100 and is not.

Let us put it in minutes, since minutes are what we live in:

Look at the middle row, because that is almost certainly where you are. Forty-three minutes a month is within the agreement. It is not a failure, you are owed nothing, nothing was breached. And yet, if those forty-three minutes land on a Friday night at a restaurant or during a campaign, they will feel like a century.

The trick is not that they lie. It is that you hear 100% and sign 99.9%. So the first practical piece of advice in this article is a boring one: read what your contract actually says, and translate it into minutes before you sign it.

And a truth you would have preferred otherwise

While researching this, I looked for any EU or Greek requirement setting a minimum availability for hosting services.

I found none. An SLA is not a regulatory floor, it is a commercial commitment. There is no law telling your provider how much of the time your site has to be up. There is only what they signed.

That changes where you look. You are not looking for compliance, you are looking at what the contract says and what happens when it is not met. And usually what happens is a credit on your next invoice, not compensation for your loss.

What you do at the moment it falls over

Here there is documentation, and it is Google’s.

If you need to take a site down, the right way is not to leave it erroring or to point it somewhere else. It is to return an informational page with a 503 status code. Google’s documentation says so explicitly for the case where you “need to urgently disable the site for 1-2 days”.

The code is not a detail for developers. It is the message “away, back soon” rather than “closed down”. With a 503, search engines know this is temporary and come back. Without it, they learn the wrong things about you.

And there is a limit worth knowing before you need it. The same page describes fully disabling a site as an extreme measure, to be taken only for a very short period, a few days at most, and warns that ramping back up will be significantly harder if the site needs reindexing.

Translation: one night of maintenance costs you nothing. Two weeks of “we will fix it next year” costs you something a button will not give back.

Do you know, or are you told?

Ask yourself something uncomfortable: if your site were down right now, how would you find out?

The two answers I usually hear are both bad. The first is “I would see it”, which means you will see it whenever you happen to open your own site, possibly not at all that day. The second is “a customer would call me”, which means the first person to find out was someone on their way to buy something.

Availability monitoring solves exactly that, and nothing else. It does not prevent the outage, nor shorten it by itself. It only changes who finds out first, and that is worth more than it sounds.

Two things need care, though, because this is where most monitoring is decorative.

The check must look at content, not just at a response. A server can answer perfectly well while the site shows a database error page, or while the certificate has expired and the browser throws a warning. Technically “it is up”. Commercially it is down. A proper check looks for a word inside the page.

Frequency defines what you can see. If the check runs every thirty minutes, a twenty-minute outage may never appear in your report. Your visitors will have lived it and you will be looking at 100%. That is not a lie, it is a blind spot.

Speed and availability, two cousins we confuse

Incidentally, “it is slow” and “it will not open” are not the same problem, though the visitor experiences them almost identically.

On speed there is one figure worth citing with its date attached: Akamai, in its spring 2017 report, stated that a delay of 100 milliseconds can hurt conversion rates by 7%. I give the year because it circulates as though it were current, and because it concerns delay, not an outage.

We have shown the relationship with our own numbers too, on osyllektis.gr and mswriter.eu. A site buckling at 19 seconds is not “down”, but the visitor does not stay long enough to learn the difference.

And now why there are no cost figures

I promised this at the start, so here is the explanation.

I went to verify the well-known figures in circulation, and they did not survive the check. The most famous, “$5,600 per minute”, is attributed to a large analyst firm but is not confirmed by that firm. The funniest finding: one of the well-known monitoring companies that reproduces the number on its own page attributes it to third parties, not to the original source. So nobody has seen the methodology, but everybody cites it.

I checked a second one that would have been perfect for this article, a supposed availability ranking of the hundred largest Greek websites. It does not exist. Nobody ever published it.

I could have written them. They would have given the article exactly the dramatic weight the subject invites, and nobody would have checked. But then what would the rest of what I told you be worth?

The cost of an outage is real. It is simply yours: it depends on what you sell, how much you sell in the hour it fell over, and how long it lasted. It is calculated on your own numbers in ten minutes, and it is more convincing than any borrowed chart.

So we close where we started, without the impressive number. The question is not what an hour of downtime costs, it is whether you will find out before your customer does and what you will do in the ten minutes after. If you want it set up properly, with a check that looks at content and not just a status code, get in touch.