Website

SLA and Uptime Explained: What 99.9 Percent Really Promises

November 11, 2026 · 7 min read · Etzal Group Team

A company signs with a cloud software provider that promises "99.9 percent uptime". Four months later the system is unavailable for three hours on the last working day of the month. Finance cannot close. The provider's reply is polite: the outage began during a scheduled maintenance window, was partly caused by a third-party network, and in any case the monthly figure is still within the commitment after exclusions. The remedy is a credit of five percent of one month's fee. Everything the provider said is consistent with the contract. Understanding SLA and uptime before signing is how you avoid learning it this way.

What an SLA is

A service level agreement states what level of service the provider commits to, how it is measured, and what happens when it is missed. A useful one has four parts:

  1. Commitments. Availability, response times, recovery.
  2. Measurement. How each is calculated, from what data.
  3. Exclusions. What does not count.
  4. Remedies. What you receive when a commitment is missed.

A sentence in a brochure saying "99.9 percent uptime" has only the first.

The percentages in hours

UptimeDowntime per 30-day monthDowntime per year
99%About 7 hours 12 minutesAbout 3 days 16 hours
99.5%About 3 hours 36 minutesAbout 1 day 20 hours
99.9%About 43 minutesAbout 8 hours 46 minutes
99.95%About 22 minutesAbout 4 hours 23 minutes
99.99%About 4 minutesAbout 53 minutes

Each extra nine costs substantially more to deliver: redundant servers, multiple locations, automatic failover, round-the-clock staff. Ask for the level the business needs, not the highest available.

Ask what the business needs

For each system, answer:

  • What does an hour of downtime cost? Lost sales, idle staff, penalties, reputation.
  • When does it matter most? Trading hours, month end, a seasonal peak.
  • How long can we manage without it? An hour with paper forms? A day?
  • How much data can we afford to lose? None, an hour, a day?

An internal reporting tool can tolerate 99 percent in business hours. An order-taking system for a distributor during trading hours needs more. A hospital system or a payment service needs far more, and a design to match.

Five questions about measurement

The percentage means nothing until you know how it is counted.

1. Over what period? Monthly or yearly. A yearly 99.9 percent permits a single outage of more than eight hours. A monthly one caps it near 43 minutes. Prefer monthly.

2. During what hours? Twenty-four hours a day, or business hours only? If your business runs 08:00 to 18:00, availability measured across 24 hours can look excellent while every outage falls in your working day.

3. What counts as "down"? Completely unreachable only? Or also so slow that it is unusable, or a key function failing while the login page still loads? A definition based on user-facing functions is stronger than one based on a server answering.

4. Who measures? The provider's own monitoring, or an independent check? Can you see the data?

5. When does the clock start? When the fault begins, when the provider detects it, or when you report it? If only on your report, unnoticed overnight outages do not count.

Common exclusions

Read this section twice. Typical exclusions:

  • Scheduled maintenance. Reasonable, if limited: how many hours a month, at what times, with how much notice.
  • Emergency maintenance. Often unlimited and unannounced.
  • Third-party failures. Network providers, cloud platforms, payment gateways. If the provider chose them, their failure is arguably the provider's risk.
  • Force majeure.
  • Customer-caused issues. Fair, when defined narrowly.
  • Features in beta.
  • Outages shorter than a threshold, such as five minutes, which may exclude frequent brief failures entirely.

Exclusions are legitimate. The question is whether, after all of them, anything meaningful remains.

Beyond uptime: the numbers that matter more

Response time. How quickly the provider acknowledges and starts working on a reported problem, by severity. For example: critical within 30 minutes at any hour, high within 2 hours, normal within one business day.

Resolution or restoration time. A target for restoring service. Providers resist firm commitments here, but a target with escalation is worth having.

Recovery time objective, RTO. After a major failure, how long until service is back.

Recovery point objective, RPO. How much data may be lost, measured in time. An RPO of one hour means backups or replication ensure you lose no more than the last hour of work.

Support hours. When can you reach a person? In which language? Through what channel?

Performance. Response time for key operations under normal load.

For many businesses, a provider with 99.5 percent availability, a 30-minute response and a one-hour RPO is a better partner than one advertising 99.99 percent with email-only support.

Severity levels

Response commitments depend on severity, so the definitions matter.

SeverityTypical definition
CriticalService unavailable, or a core function unusable for all users, no workaround
HighA major function impaired, or many users affected, workaround difficult
MediumA function impaired with a reasonable workaround
LowMinor issue, question or request

Agree who assigns severity, and how a disagreement is escalated.

Remedies

Service credits. A percentage of the fee, increasing with the size of the miss. Check:

  • Are credits automatic, or must you claim them within a deadline?
  • Is there a cap?
  • Are they a credit against future fees, or a refund?

Termination right. The ability to leave without penalty after repeated or severe failures. This is often worth more than credits.

Root cause report. A written explanation after a serious incident: what happened, why, and what will change.

Credits will not compensate for business loss. Do not rely on an SLA as insurance.

Five terms worth negotiating

Standard terms are a starting point. For a system that matters, ask for:

  1. Monthly measurement during your business hours, or a higher commitment in those hours.
  2. A definition of downtime that includes key functions and severe slowness.
  3. Limits on maintenance: maximum hours per month, outside your working hours, with notice.
  4. Response times by severity, with named escalation contacts.
  5. A termination right after a set number of missed months, with assistance to move your data.

Large public cloud services rarely negotiate. Smaller providers and custom software vendors usually do.

For custom-built systems

When a vendor builds and operates a system for you, the SLA is designed, not inherited.

  • Availability depends on architecture. A single server cannot deliver 99.99 percent, whatever the contract says.
  • State the target early, because it determines cost: redundancy, monitoring, on-call staff.
  • Separate the hosting provider's commitment from the vendor's. The vendor's promise cannot exceed what the infrastructure supports unless redundancy is designed in.
  • Agree backup frequency, restore tests, RTO and RPO as part of the design.

Your own side of the bargain

An SLA does not remove your responsibilities.

  • Monitor independently. A simple external check tells you about an outage when it starts, and gives you your own record.
  • Keep your own backups or exports of critical data where the service allows.
  • Have a manual fallback for the processes that cannot stop.
  • Know how to reach support at 02:00, and test it once.
  • Review the provider's performance quarterly against the commitments.

Common mistakes

  • Reading only the percentage.
  • Not checking the measurement period and hours.
  • Ignoring exclusions.
  • No response time commitment.
  • Assuming credits cover losses.
  • Demanding the highest uptime without needing it, and paying for it.
  • No independent monitoring.
  • No exit right.

We set availability targets with clients during design and build the monitoring and recovery to match, as part of Enterprise Software. If you are reviewing an SLA and want a second reading, contact us.

Summary

Translate the percentage into hours and compare it with what an outage costs you and when. Check how availability is measured: period, hours, definition of down, who measures and when the clock starts. Read the exclusions. Give more weight to response time, restoration, RTO and RPO than to the headline figure. Negotiate business-hours measurement, maintenance limits, severity-based response and a termination right, and keep your own monitoring, backups and fallback.

Frequently asked questions

How much downtime does 99.9 percent uptime allow?

About 43 minutes in a 30-day month, or roughly 8 hours and 46 minutes in a year. Whether that is acceptable depends on when it falls and how it is counted. Forty minutes at 03:00 on a Sunday and forty minutes during month-end closing are very different events.

Is uptime the most important number in an SLA?

Often not. For most business systems, how quickly the provider responds and restores service, how much data can be lost, and what counts as downtime matter more than the headline percentage. A high uptime figure with broad exclusions and no response commitment offers little protection.

Do service credits compensate for the cost of an outage?

Rarely. Credits are usually a percentage of the monthly fee, which is far less than the business loses in an outage. Treat credits as an incentive for the provider and a signal of seriousness. Protection against real loss comes from architecture, backups, and your own continuity plan.

Author

Etzal Group Team

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

Website Development

Fast, SEO-ready websites that turn visitors into customers.