OpusBUSINESS EXPERT

99% and 99.9% look like the same promise. One of them is three and a half days a year

Andy Shepherd7 min read
  • Technology
  • IT consultancy
  • Hosting
  • Working with us
  • Knutsford

Two quotes on the desk. One promises 99.9% availability, the other promises 99%, and the second one is cheaper. The difference reads like a rounding error.

Over a year it is seventy-nine hours.

87.6 hrs
99% allows, over a year
8.8 hrs
99.9% allows, over a year
43 mins
99.9% allows, in a 30-day month

That is the whole of the arithmetic, and it is worth doing once because the numbers are counter-intuitive. A tenth of a per cent sounds like nothing. A whole per cent sounds like nearly nothing. One of them is three and a half days with the shutters down.

The percentage is also close to the least important thing in the clause. Four other things decide what it is actually worth, and a supplier who wants to look good has latitude in every one of them.

This is the availability clock. The response clock in the same contract is a separate promise with its own settings, and it is the one most firms end up caring about more.

The window decides what the number means

An availability target is measured over a period, and the period is doing at least as much work as the figure.

Measured monthly, 99.9% is forty-three minutes. Twelve separate allowances, each of which has to be met on its own, so a single four-hour outage in March breaks March and triggers whatever the remedy is.

Measured annually, the same 99.9% is eight and three-quarter hours in one pot. That same four-hour outage uses half the year's allowance and breaches nothing at all. The supplier has met the target. You lost an afternoon.

The two windows hide opposite failures. An annual window hides the one catastrophic day. A monthly window hides the slow drip: forty minutes every single month, every month, is eight hours a year of being offline inside a contract nobody has ever breached.

Neither is dishonest. But if you are comparing two quotes and only one of them states its window, you are not comparing like with like, and the one that leaves it out is not the one being careless.

Most of what takes a small firm offline is excluded

Planned maintenance is nearly always outside the calculation. That is reasonable in itself, and the question is what counts as planned, how much notice you get and whether there is any cap on how much of it there can be.

The bigger exclusion is everything upstream. A supplier can only promise what they control, so the clause carves out your broadband, your registrar, your certificate, the cloud platform they are reselling and whatever your staff are doing at home. Read the list and you find that a decent share of the ways a ten-person firm actually ends up unable to work are, correctly and explicitly, not covered by the uptime guarantee it is paying for.

Which is the real reason the percentage is a poor thing to shop on. It measures a slice of the problem. What still works when the broadband goes down is the other slice, and no hosting contract will ever cover it.

"Down" is a definition, not an observation

Most agreements count an outage when the service as a whole is unreachable. That is not usually how it goes wrong.

What happens is that the website loads but the contact form does not send, or the system is fine except for the one report the finance meeting needs, or everything works and nothing is slow enough to call broken. By a strict whole-service definition, none of those is downtime. Your morning disagrees.

So ask how partial failure is treated, and ask who declares it. If the answer is that the supplier's monitoring decides, then the supplier is marking their own homework, and you should at least know the exam is theirs.

The remedy is a refund of what you paid, not of what it cost

This is the part that surprises people most. The standard remedy for a missed availability target is a service credit: a percentage off the next invoice, scaled to how badly the target was missed.

Put numbers on it. If the service costs a hundred pounds a month and the contract offers a 10% credit for that tier of breach, then a day in which your firm could not invoice, quote or answer email is worth ten pounds. The credit is sized to the supplier's revenue from you, which has no relationship at all to what the outage did to your week.

Whether a credit-only remedy can be the limit of what you are owed is a question for a solicitor, not for us or for a blog post. The practical point stands either way: the clause is not insurance, and if the cost of a bad day would genuinely hurt, the place to deal with that is your own insurance and your own fallback plan rather than a line in a hosting agreement.

We have had to decide where these lines go

We are not reading these contracts from the outside. We host client sites, and we have moved a client off an ageing phone system onto Teams Phone with their numbers ported, so we carry availability as well as writing about it.

More to the point, we built the clocks. Opus Help Desk, which we wrote for a payroll partner, monitors service levels with ageing counted on working days only, and escalates on a ladder that changes by time of day and day of week. Every one of those rules had to be chosen deliberately, and each had a version that would have flattered whoever was being measured. Counting ageing on working days is the honest reading of a rule that could just as easily have been written to make Friday afternoon disappear.

That is why we are unimpressed by percentages. We know how much of the result lives in the definitions underneath.

What to negotiate instead

Not a bigger number. Pushing 99.5% up to 99.9% is the concession a supplier can make most cheaply, because the definitions around it stay exactly where they were.

Ask for these four in writing before you sign, and keep the answers with the contract:

  1. 01Is the window a month or a year, and when does the allowance reset
  2. 02What is excluded, and how much planned maintenance is permitted
  3. 03Does a single broken feature count as downtime, and who declares it
  4. 04What is the target time to fix, and who do I ring when that target is passing

The fourth one is the one that matters on the day. Availability tells you how often you can expect to be broken. Time to fix, and a named escalation route when the fix is not happening, is the only part of the clause you will care about while it is happening.

If you have two quotes and cannot tell which is the better deal, the percentages are not where the difference is. Send us both and we will read the definitions. Hosting and infrastructure is one of the things we do, so treat that offer with the appropriate suspicion and take us up on it anyway.


We work with businesses across Knutsford, Alderley Edge, Wilmslow, Altrincham, Stockport and Warrington, and remotely for clients anywhere in the UK.

If any of this sounds like your business, we will tell you plainly whether we can help.

We would like to use analytics cookies to understand how this site is used, advertising cookies to measure our ads, and to load our office map from Google. None of them loads on its own unless you accept — though Google Tag Manager itself loads either way, with every category switched off. The map also has a button of its own. Read our cookie policy