Latest Insights/Back to Generator
By the Legal Policy Generator team · Published 2026-11-13

What Is an SLA? What a Service Level Agreement Must Include

99.9%43.8 min / month

A service level agreement (SLA) is the part of a B2B contract that turns "we'll keep the service running" into numbers. It states how much uptime the provider commits to, how fast support responds, how both are measured, and what the customer gets back when the provider misses. If an enterprise buyer has just asked for one, the free SLA Generator drafts a complete agreement with uptime tiers, response times and service credits in a few minutes, no signup needed.

Key takeaways

  • An SLA is a measurement contract. It defines a handful of metrics, the target for each, how they're measured, and the remedy when a target is missed. Everything else is supporting detail.
  • Seven clauses do the work: scope, uptime commitment, measurement method, support response and resolution times, service credits, exclusions, and the claims process.
  • "Three nines" still allows real downtime. 99.9% uptime permits 43.8 minutes of downtime in an average month, and about 8 hours 46 minutes over a year.
  • Credits cap your exposure. Write service credits in as the customer's sole and exclusive remedy for missed targets. That one clause is what turns credits into a liability ceiling.

What is an SLA (service level agreement)?

An SLA is a contract, or a schedule attached to one, that sets measurable service levels between a provider and a customer. Where terms of service describe what the service is and the rules for using it, the SLA describes how reliably it will be delivered and what happens when it isn't. The two documents are usually read together: the terms say "the service is provided as described", and the SLA says "and here is the uptime and support we commit to, measured this way".

People swap three terms constantly. Here is how this guide uses them:

  • SLI (service level indicator) is the thing you measure. Monthly uptime percentage, first-response time on a support ticket, or API error rate.
  • SLO (service level objective) is the internal target your engineering team aims for. Keep it stricter than anything in the contract.
  • SLA (service level agreement) is the external commitment you put in front of a customer, with a remedy attached. Miss the SLO and you have an engineering conversation. Miss the SLA and you owe someone a credit.

A practical rule: set the SLA a notch below the SLO. If your systems reliably hit 99.95%, committing to 99.9% in the contract gives you headroom for the one bad month.

Who needs one

Consumer products rarely need one, and a small B2B tool can usually get by without one until a customer asks. Three situations change that:

  • Enterprise procurement. Larger buyers often run vendors through a security and legal questionnaire, and "do you offer an SLA?" tends to be on it. Treat a missing SLA as a reason the deal stalls, not something to negotiate around later.
  • You run something other businesses depend on in real time. Payments, hosting, authentication, a CRM the sales team lives in, managed IT services. If your outage becomes their outage, they'll want your commitment in writing.
  • You want to limit your own liability. This is the underrated half. Without an SLA, a customer who loses a day of sales during your outage argues for actual damages. With one, the agreed credit is the remedy and the argument is over before it starts.

If you're selling to businesses and have reached the stage where contracts are being redlined, an SLA belongs on the same checklist as your data processing agreement and acceptable use policy. The SaaS launch legal checklist shows where it sits among the rest.

The seven clauses every SLA needs

1. Scope: which services, which customers, which plan

Say exactly what the SLA covers. "The hosted application at app.example.com" is specific. "The Services" is not, unless the parent agreement defines it. If you offer the SLA only on paid or higher-tier plans, say so here. If a free tier or beta feature is excluded, this is the clause that excludes it.

2. The uptime commitment

The headline number. State it as a percentage per measurement period: "Monthly Uptime Percentage of at least 99.9%". Then define the term, because "uptime" on its own is ambiguous. Does a slow response count as up? Does a degraded feature count as down? One workable definition treats the service as down only when it's unavailable for all users for a continuous period of at least a few minutes, and treats partial degradation separately or not at all. Whatever you choose, write it down.

3. How uptime is measured

Start with the formula, because every other detail hangs off it:

Monthly Uptime Percentage = (total minutes in the month − minutes of qualifying downtime) ÷ total minutes in the month × 100

Worked example: 90 minutes of qualifying downtime in a 30-day month (43,200 minutes) gives (43,200 − 90) ÷ 43,200 × 100 = 99.79%. Against a 99.9% commitment that month is a miss, and in the credit schedule in clause 5 below it lands in the 10% tier. (The downtime table later in this guide uses a 30.4375-day average month instead of a flat 30 days.)

Skipped details get argued over later:

  • The measurement period. Monthly is the customer-friendly choice. Quarterly or annual windows let a provider absorb one long incident, as the math below shows.
  • Who measures. The provider's own monitoring, a named third-party monitoring service, or the customer's logs. If you rely on your own monitoring, say so.
  • Minimum incident length. You can choose not to count a 30-second blip. State the threshold (for instance, five consecutive minutes) so both sides know what qualifies.

4. Support response and resolution times

Uptime is only half of what buyers want. The other half is "when something breaks, how fast do you answer?" Answer it with a severity table: a response target for each level, and optionally a resolution or workaround target too. Here is an example structure for B2B software:

Severity Meaning Example first-response target Hours covered
Severity 1Service down or unusable for all users, no workaround1 hour24/7
Severity 2Major feature broken, business impact, workaround exists4 hours24/7 or extended business hours
Severity 3Minor defect, limited impact1 business dayBusiness hours
Severity 4Question, cosmetic issue, feature request3 business daysBusiness hours

Define "business hours" with a time zone, because your customer's 9 to 5 may not be yours. And commit to response times freely but to resolution times carefully: you control how fast someone replies, you don't always control how fast a bug can be fixed.

5. Service credits

The remedy. The usual form is a credit rather than a refund: a percentage of that month's fee applied against a future invoice. Tier the credits so they grow as uptime falls further below the target. An illustrative schedule for a 99.9% commitment:

Monthly uptime actually delivered Credit (% of that month's fee) On a $1,000/month plan
99.9% or aboveNone (target met)$0
Below 99.9%, at or above 99.0%10%$100
Below 99.0%, at or above 95.0%25%$250
Below 95.0%50%$500

Put a cap and an exclusive-remedy clause next to the table. The cap says total credits in any month can't exceed some share of that month's fee (the 50% top tier above acts as the cap). The sole and exclusive remedy wording makes credits the customer's only compensation for the SLA failure. Sample language:

Service Credits are the Customer's sole and exclusive remedy for any failure by the Provider to meet the Service Levels in this Agreement. The total Service Credits in any calendar month will not exceed 50% of the fees payable for that month.

Those two sentences are what turn an SLA from a marketing promise into liability protection, subject to the governing law and to any carve-outs the parties agree. Expect enterprise buyers to ask for one carve-out in particular: a right to terminate for chronic failure, such as missing the uptime target in three consecutive months. Decide before negotiation whether you'll grant it. Pair the whole clause with the limitation-of-liability section in your SaaS terms of service so the two don't contradict each other.

6. Exclusions

Downtime that doesn't count against the commitment. The usual candidates:

  • Scheduled maintenance, with a notice period (say, 48 hours) and a cap on how many hours a month qualify. Unlimited "maintenance" swallows the whole SLA, so expect buyers to strike it.
  • Force majeure: events outside anyone's reasonable control.
  • Customer-caused outages: misconfiguration, exceeding agreed usage limits, or problems in the customer's own network or equipment.
  • Third-party failures the provider doesn't control, such as the customer's internet service provider or a domain registrar. Be careful here: if you host on a cloud platform, a buyer may push back on excluding your infrastructure vendor's outages, since choosing that vendor was your decision.
  • Beta, preview or free features, which you can offer without any service level.

7. The claims process

How the customer actually gets a credit. Require the request in writing within a set window, for instance 30 days after the month in question, with enough detail to identify the incident. Say that unclaimed credits are forfeited, that credits have no cash value, and that they don't survive termination. Spell all of this out, because an SLA without a claims process produces disputes about whether a credit was "owed automatically".

What 99.9% uptime actually allows

Uptime percentages are easier to negotiate once you translate them into minutes. The figures below use an average month of 30.4375 days (43,830 minutes) and a year of 365.25 days (525,960 minutes), so a 31-day month allows slightly more and February slightly less.

Uptime commitment Nickname Downtime allowed per month Downtime allowed per year
99%Two nines7 h 18 min3 d 15 h 40 min
99.5%n/a3 h 39 min1 d 19 h 50 min
99.9%Three nines43.8 min8 h 46 min
99.95%n/a21.9 min4 h 23 min
99.99%Four nines4.4 min52.6 min
99.999%Five nines26 seconds5.3 min
Downtime allowed per average month by uptime commitmentDowntime allowed per average month (minutes)99%43899.5%21999.9%43.899.95%21.999.99%4.4Basis: 30.4375-day average month = 43,830 minutes. Bars scaled to 438 min = full width.

Look at the step from 99.9% to 99.99%: it cuts the allowance tenfold, from about 44 minutes to about 4 minutes a month. Four nines is hard to deliver for a small team, so commit to it only if your monitoring history supports it. The measurement window matters as much as the percentage. A 99.9% commitment measured quarterly allows roughly 131 minutes across the quarter, so a single two-hour incident can pass while the same incident would breach a monthly SLA by a wide margin. Buyers should push for monthly measurement. Providers should know what they're giving up when they agree to it.

Common mistakes

  • Promising a number engineering hasn't hit. Look at the last twelve months of real uptime before writing the commitment. If you had one 90-minute incident this year, 99.99% is not your number.
  • Leaving "downtime" undefined. If the contract doesn't say whether slow responses or partial outages count, the customer will say they do and you will say they don't.
  • Open-ended maintenance exclusions. Cap the hours and require notice, or expect the clause to be struck in negotiation.
  • Committing to resolution times you can't control. Respond-within-one-hour is a staffing promise you can keep. Fix-within-four-hours depends on the bug.
  • No "sole and exclusive remedy" sentence. Without it, the credit schedule is a floor on what you owe, not a ceiling.
  • Letting the SLA and the terms disagree. If the terms disclaim all liability and the SLA promises credits, say which document wins for SLA matters. Usually the SLA governs its own subject and the terms govern everything else.

Where to put it

The simplest approach is to publish the SLA as a standalone page linked from the terms of service and from the pricing or enterprise page, so sales can send a URL instead of a PDF. Larger cloud vendors do exactly this. Google Cloud publishes its service level agreements as public pages, and Microsoft maintains SLA documents for its online services covering uptime and connectivity commitments. Reading a few of them is the fastest way to see how uptime definitions, exclusions and credit tiers are worded in practice. For a negotiated enterprise contract, the same SLA is attached as a schedule and referenced by clause number from the master agreement.

The policy guide lists the other documents a B2B service usually needs around it, and the acceptable use policy handles the mirror-image question of what the customer commits to.

Frequently asked questions

Is an SLA legally binding?

Yes, when it's part of a signed contract or incorporated by reference into terms the customer accepted. A published SLA page that the terms point to is binding in the same way the terms are. A marketing page that says "99.9% uptime" with no contractual language is a claim, not an SLA.

What's the difference between an SLA and a warranty?

A warranty is a general promise that the service will perform as described, with ordinary contract remedies if it doesn't. An SLA is narrower and more mechanical: specific metrics, specific targets, and a pre-agreed remedy (credits) that usually replaces those ordinary remedies for the metrics it covers.

Should a startup offer an SLA?

Only once customers ask, and then only at a level the team can actually sustain. A conservative 99.5% monthly commitment you always meet is worth more than a 99.99% commitment you miss twice a year. Keeping the SLA to business and enterprise plans, and leaving self-serve tiers without one, is a reasonable way to start.

Do service credits have cash value?

Not unless the agreement says so. Provider-friendly drafting limits credits to future invoices, bars exchanging them for cash, and ends them when the agreement ends. If a customer is on their final month, that limitation matters, and buyers sometimes negotiate a cash-out for credits earned in the last billing period.

Can I use one SLA for every customer?

For a self-serve or mid-market product, yes: one published SLA keeps things simple. Enterprise deals may bring negotiated variations, such as a stricter uptime target or a higher credit cap. Keep those as a short addendum that overrides specific clauses rather than a separate full document, so you're maintaining one base text.

Related reading

Draft your SLA with a free SLA generator

Ready to draft yours? Set your uptime tier, support response times and credit schedule in the free SLA Generator and download the finished agreement as PDF or HTML. No signup required.

This article is general information about a legal topic, not legal advice for any specific situation. Laws change, jurisdictions vary, and the right answer depends on facts a generator can't see. Consult a licensed attorney in your jurisdiction for advice specific to your business.