Global perspectives · Ideas that move business forwardRSS

Tech / SaaS / Business
Ideas · Innovation · Impact

Global perspectives
for a smarter tomorrow.

SaaS

How SaaS Uptime Guarantees and SLA Credits Work

A 99.9% uptime promise sounds nearly perfect, but it still allows hours of downtime a year. Learn how to read a service level agreement, do the math and plan around the gaps.

Graphs of performance analytics on a laptop screen

Almost every SaaS provider advertises an uptime percentage, and almost every buyer reads it as a promise that the product will simply work. This article explains what those percentages actually permit, what a service level agreement can and cannot do for you, and how to plan for the gaps.

What is an SLA, and why does a SaaS company publish one?

A service level agreement, or SLA, is the part of a contract or public policy in which a provider states how well its service will perform and what happens if it falls short. For software delivered over the internet, the most common commitment is availability: the share of time the service is working and reachable.

SaaS is a natural home for these promises. The US National Institute of Standards and Technology (NIST) describes cloud computing as a model for “ubiquitous, convenient, on-demand network access” to shared computing resources, which can be provisioned and released with minimal management effort. When customers rent software that way rather than run it themselves, they hand over the job of keeping it running. An SLA is the written record of how well that job will be done.

The UK’s National Cyber Security Centre (NCSC) notes in its cloud security guidance that providers will likely use contractual commitments or SLAs to state their availability promises. In other words, the SLA is the main formal assurance a customer gets. It is worth reading as carefully as the price.

What does 99.9% uptime actually allow?

Availability is usually written as a percentage with a number of nines. The percentage looks tiny when it is under 100, but the allowed downtime is easy to calculate. A 30-day month has 43,200 minutes (30 days × 24 hours × 60 minutes). A 365-day year has 525,600 minutes.

Promised availability Downtime allowed in a 30-day month Downtime allowed in a year
99% 432 minutes (7.2 hours) About 87.6 hours
99.9% 43.2 minutes 525.6 minutes (8.76 hours)
99.95% 21.6 minutes 262.8 minutes (4.38 hours)
99.99% 4.32 minutes 52.56 minutes

This is illustrative arithmetic, not any particular vendor’s terms. The point is the scale of the differences. Moving from 99.9% to 99.99% cuts allowed downtime by a factor of ten, which usually requires far more engineering and costs the provider far more. That cost often shows up in higher tiers, which is one reason availability promises sit alongside the pricing structures covered in how SaaS pricing models work.

How is uptime measured, and what is left out?

The percentage is only as meaningful as the definition behind it. Several terms in the fine print decide whether an outage “counts.”

  • Measurement window. A monthly window is stricter than an annual one. With a yearly window, one long outage can be averaged against eleven quiet months.
  • Excluded downtime. Many SLAs exclude scheduled maintenance, events outside the provider’s control, and problems caused by the customer’s own configuration or network.
  • Definition of “down.” Some agreements count only a complete failure. A product that loads but cannot save data, or is so slow it is unusable, may not qualify.
  • Who measures. If the provider’s own monitoring is the only record, the customer may need to file a claim and supply evidence.
  • Minimum duration. Brief interruptions below a stated threshold may be ignored.

The NCSC adds a structural point: providers usually define their SLAs per service, rather than for the whole of their platform. A vendor may therefore guarantee its core application but make no promise about an add-on, an API or a reporting feature that your team depends on daily.

How do SLA credits work, and what do they really compensate?

When a provider misses its target, the remedy is typically a service credit: a discount on a future invoice, often a percentage of that month’s fee that rises as availability falls. Credits usually have to be requested within a set period, and many agreements cap them at a share of one month’s fees. Terms differ widely, so these are general patterns rather than rules.

A worked example shows why credits rarely cover real losses. Suppose a team pays $2,000 a month for a tool, the SLA promises 99.9%, and a month’s availability turns out to be 99.0%. That is 432 minutes of downtime against an allowance of 43.2. If the agreement gives a 10% credit at that level, the credit is $200. If the outage stopped the team from invoicing customers for a day, the business cost could easily exceed $200.

The NCSC makes the same observation in general terms: an SLA may provide a mechanism for compensation if outages occur, but outages will not be prevented if the service design is not appropriate. A credit is a pricing adjustment, not insurance. It also does not repair the harm of lost trust, which matters because reliability problems feed directly into the cancellation patterns described in SaaS churn explained.

Why can two reliable services still produce an unreliable workflow?

Availability numbers compound. If a workflow needs two services to work at once, and each is available 99.9% of the time, the combined availability is roughly 0.999 × 0.999 = 0.998001, or about 99.8%. Over a 30-day month, that is roughly 86 minutes of expected downtime rather than 43, assuming the two fail independently.

Modern products often chain together a sign-in provider, a payment processor, a messaging service and the application itself. Each link has its own SLA, and the customer experiences the product of all of them.

The NCSC’s SaaS guidance flags one common example: federated sign-in, where one identity provider logs users into many applications, “introduces a large single point of failure.” If that provider is unavailable, every application behind it may be unreachable, even if each one is running perfectly.

Redundancy works in the opposite direction. If two independent components each have 99.9% availability and either one can carry the load, the chance both fail together is 0.001 × 0.001 = 0.000001, which implies about 99.9999% availability under ideal assumptions. Real systems fall short of that because failures are often correlated and failover mechanisms can themselves fail.

What can builders do when the vendor’s number is not enough?

If you are building a SaaS product, you face the same problem from the other side: your customers will rely on your promise, and your own suppliers’ promises limit it. A product cannot sensibly promise higher availability than the weakest service it depends on, unless it is designed around that weakness.

The NCSC suggests that when a provider does not offer the availability percentage you need, you can often still reach your target through redundancy, for example by using a high-availability architecture across multiple data centers or geographic regions. It also cautions that ineffective failover decreases availability, so failover should be tested regularly rather than assumed to work.

Practical steps for builders include:

  1. Map every dependency a customer-facing feature needs, and note each one’s stated availability.
  2. Set an internal target stricter than the public promise, so there is a safety margin.
  3. Publish an honest SLA with clear definitions, and a public status page. The NCSC observes that a publicly available SLA can show a higher commitment to meeting claims than one found only in a contract.
  4. Decide in advance how credits are requested or applied automatically, so a bad month does not become a support burden on top of an engineering one.
  5. Rehearse failures, including the recovery path, before they happen in production.

What should buyers check before relying on an availability promise?

Buyers can treat the SLA as one input within a broader evaluation, alongside the criteria in how to evaluate a SaaS tool before you commit. A short checklist helps:

  • Scope. Does the SLA cover the specific features you depend on, or only the core product?
  • Definitions and exclusions. How is downtime defined, and what is excluded?
  • Window and remedy. Is availability measured monthly, and what are the credit tiers and caps?
  • Support commitments. The NCSC advises that services bought without a contractual commitment for support should be considered to have no guaranteed support. Check response times as well as uptime.
  • Track record. The provider may share historical availability data. The NCSC suggests weighing it alongside the provider’s commitments and reputation, since past performance is evidence, not a guarantee.
  • Incident communication. The NCSC recommends becoming familiar with how the application reports outages and incidents, such as through a status page.

Which resilience tasks stay with the customer?

Even a strong SLA leaves work on the customer’s side. The NCSC’s SaaS guidance says customers remain responsible for configuration specific to their use of the application, and it advises verifying that critical data in the application is stored in at least one resilient backup. Backups should not share the same failure points: an incident affecting the application’s data should not be able to affect all the backups too.

The guidance also recommends access recovery procedures, sometimes called emergency or break-glass access, so that administrators can regain control if normal sign-in fails during an incident. These tasks sit within the wider split of duties described in the SaaS shared responsibility security checklist.

Reading the nines with open eyes

An uptime percentage is a useful summary, but it is a promise about averages inside a defined window, with exclusions, and backed by a remedy that is usually a bill discount. For buyers, the sensible response is to read the definitions, compare the allowed downtime in minutes rather than nines, and plan for the dependencies that sit outside any single contract. For builders, it is to promise only what the whole dependency chain can deliver and to test the recovery path before customers do.

This article is general information about software contracts and reliability, not legal or financial advice. Check the actual terms of any agreement you sign.

Sources: National Cyber Security Centre; National Institute of Standards and Technology