Sit down for a minute, because there is a small piece of arithmetic that will save you from a lot of awkward moments in front of customers. A buyer’s IT head says, “Your SLA is 99.9 per cent, right?” and you nod. Do you know what you just agreed to? Let us work it out together.

Availability is the share of time a service is working. Take a 30-day month. That is 30 × 24 × 60 = 43,200 minutes. If the promise is 99.9 per cent, the service is allowed to be unavailable for 0.1 per cent of that, which is 43.2 minutes. Not an hour, not a day. Forty-odd minutes in the whole month.

Now do the same for the other numbers you will hear. For a 30-day month, 99.5 per cent allows 216 minutes, which is three and a half hours. 99.95 per cent allows 21.6 minutes. 99.99 per cent, the famous “four nines”, allows about 4.3 minutes. Over a full year, 99.9 per cent is roughly 8.76 hours of downtime, and 99.99 per cent is under 53 minutes. Each extra nine cuts the allowed downtime to a tenth, and it usually costs the provider far more than ten times the engineering effort to deliver it.

Same percentage, different promise

Here is where it gets interesting. Two vendors can both write “99.9%” and mean quite different things. You should ask four questions about any SLA before you quote it.

  • Over what window is it measured? Monthly, quarterly or annual. A yearly 99.9 per cent allows one bad 8-hour day and still passes. A monthly one does not.
  • What counts as “down”? Complete unavailability only, or also severe slowness, or failure of one feature such as report export? Many contracts count only the core service responding.
  • What is excluded? Scheduled maintenance windows, customer-caused issues, failures in the customer’s own network, and events outside the provider’s control are commonly carved out. If maintenance is excluded and the vendor schedules eight hours a month, the headline number hides a lot.
  • Who measures? The vendor’s own monitoring, or a third party, or the customer’s logs? In a dispute this decides everything.

Notice that none of these questions is technical. They are contract questions, and the salesperson is often the only person in the room who has read the draft.

What happens when the vendor misses

An SLA with no consequence is a wish. The usual remedy is a service credit: a percentage of that month’s fee returned or set against the next invoice. A typical structure has tiers, for example a small credit if availability falls below 99.9 per cent, a bigger one below 99.5 per cent, and so on. The exact numbers vary hugely between vendors and contracts, so treat that example as a shape, not a norm.

Let us make it concrete. Suppose a customer pays ₹2 lakh a month and one month had 130 minutes of qualifying downtime. Against 43,200 minutes, that is 130 ÷ 43,200 = about 0.30 per cent downtime, so measured availability is roughly 99.7 per cent. It missed 99.9 per cent but beat 99.5 per cent. If the contract gives a 10 per cent credit in that band, the customer gets ₹20,000 back. Compare that with what an outage of two hours might have cost the customer’s own business on a busy day, and you see why large buyers push for higher credits, termination rights after repeated misses, and not just a refund.

Two details trip people up. First, credits are often not automatic; the customer must claim within a set number of days. Second, credits are very often the sole remedy, which means a serious outage costs the vendor only a slice of one month’s fee. Buyers in regulated sectors tend to negotiate this harder than anyone else.

A percentage is not a recovery story

Here is a thought that a thoughtful buyer will raise, so raise it first. Ten outages of four minutes each and one outage of forty minutes both produce 99.9 per cent. For a payments page those are very different experiences. Ten small blips may be fine for a back-office tool and terrible for something customers use live.

That is why serious buyers ask for other measures next to availability. Support response and resolution targets by severity, how quickly the vendor will tell them about an incident, the target for restoring service (often written as RTO) and how much recent data could be lost (RPO). Ask about a published status page and post-incident reports too. A vendor that explains its last outage plainly is usually more trustworthy than one that quotes five nines and goes quiet when asked about the past year.

What to do with this on your next call

Do not promise a number your engineering team has not agreed to, and do not let a customer talk you into “100 per cent” in a contract, because no real service delivers it. If the buyer asks for more nines than your standard, check whether the cost to you is justified by the deal, because a higher tier may need redundant regions or extra staff on call and that belongs in the price. If you are on the buying side, convert every percentage into minutes first and then read the exclusions.

Also remember that your SLA is only as strong as your dependencies. If you run on a cloud provider that itself promises a certain figure, your own promise cannot sensibly exceed what you can build on top of it unless you design around failure. Ask your architect, not your gut.

Figures and names in the examples are illustrative.

Leave a Reply

Trending