---
title: "How to Write a Vendor SLA: Key Metrics, Service Credits and Review Terms"
url: https://businesslang.com/guides/how-to-write-a-vendor-sla/
published: 2026-11-13
updated: 2026-11-13
---

# How to Write a Vendor SLA: Key Metrics, Service Credits and Review Terms

> Write a clear vendor SLA: which metrics to set, how uptime math and service credits work, what Google, Microsoft and AWS SLAs promise, plus sample clauses.

- URL: https://businesslang.com/guides/how-to-write-a-vendor-sla/
- Type: Buying Guides
- Category: Software Buying
- Author: Aram Petrosyan, Senior Procurement Manager; Alessandro Romano, Senior Business and Procurement Consultant
- Published: 2026-11-13
- Last updated: 2026-11-13
- Topics: sla, service-level-agreement, vendor-management, uptime, service-credits, saas-contracts, vendor-performance, negotiation
- Cite as: Aram Petrosyan and Alessandro Romano, "How to Write a Vendor SLA: Key Metrics, Service Credits and Review Terms", Businesslang, November 13, 2026, https://businesslang.com/guides/how-to-write-a-vendor-sla/

**Short answer:** A clear vendor SLA defines each service you depend on, sets measurable targets (uptime, response and resolution time by severity, recovery objectives, support hours or on-time delivery), explains exactly how each target is measured, lists a short set of exclusions, and states what happens when a target is missed: service credits, how to claim them, any cap, and a right to terminate after repeated failures. It should also require regular performance reports and a scheduled review. Big SaaS vendors publish standard SLAs, usually 99.9% monthly uptime with credits as the sole remedy, so use those as a baseline and negotiate the gaps that matter to your business.

A service level agreement (SLA) is where a vendor's promises become measurable. "Highly available" and "fast support" mean nothing on the day your order system is down; "99.9% monthly uptime, a 1 hour response to severity 1 tickets and a 25% credit if either is missed" is something you can check, claim and use at renewal.

Most small businesses accept the vendor's standard SLA, or sign with no service levels at all. That is fine for low-risk tools. For the few vendors you cannot run without, such as your accounting system, payment provider, managed IT partner or main supplier, a clear SLA is worth an afternoon of work. This guide covers the metrics, the uptime math, credits and termination rights, what Google, Microsoft, AWS and Slack actually promise, and a drafting process with sample clauses. For the rest of the contract, see our [SaaS contract checklist](https://businesslang.com/guides/saas-contract-checklist/).

## What an SLA is, and how it differs from the contract

An SLA defines the level of service a vendor commits to deliver, how that level is measured and what happens if it is missed. It usually sits inside a wider set of documents:

| Document | What it covers | Typical owner on your side |
|---|---|---|
| Master services agreement (MSA) or terms of service | The legal relationship: liability, warranties, confidentiality, governing law, termination | Owner or legal counsel |
| Order form or statement of work (SOW) | What you are buying, quantities, price, term and deliverables | Buyer or budget owner |
| Service level agreement (SLA) | Measurable targets, measurement method, exclusions, credits, reporting and review | Operations or IT lead |
| Data processing agreement (DPA) and security terms | How personal data is handled, breach notice, audits | IT or compliance lead |

Make sure the contract incorporates the SLA by name and version, since web-published SLAs change; Google's SLA page lists nine previous versions. Also check which document wins in a conflict. AWS's EC2 SLA says that where it conflicts with the AWS Customer Agreement, the SLA terms apply "only to the extent of such conflict."

## What real published SLAs promise

Before drafting your own terms, look at what large vendors commit to as standard. These figures come from each vendor's SLA page, checked on November 13, 2026.

| Vendor and service | Uptime commitment | Service credit if missed | Claim deadline | Remedy |
|---|---|---|---|---|
| Google Workspace (Gmail, Drive, Meet and other covered services) | At least 99.9% Monthly Uptime Percentage | 3 days of service below 99.9%; 7 days below 99.0%; 15 days below 95.0% | Within 30 days of becoming eligible | Credits are the "sole and exclusive remedy"; maximum 15 days per month |
| Microsoft Online Services (Exchange Online, Teams, SharePoint Online) | 99.9% uptime | 25% below 99.9%; 50% below 99%; 100% below 95% | Example given: incident on February 15, claim due by March 31 | Credits are the "sole and exclusive remedy"; capped at the monthly service fee |
| AWS EC2, Region-Level | 99.99% Monthly Uptime Percentage, "commercially reasonable efforts" | 10% below 99.99%; 30% below 99.0%; 100% below 95.0% | By the end of the second billing cycle after the incident | Credits apply only to future EC2 payments, issued only if over $1 |
| AWS EC2, Instance-Level | 99.5% per instance | 10% below 99.5%; 30% below 99.0%; 100% below 95.0% | Same as above | Same as above |
| Slack | "Commercially reasonable efforts" to be available 24/7, excluding planned downtime and events beyond its control | None stated on the page | None stated | None stated |

The lessons:

- **99.9% is the common baseline** for business SaaS.
- **Credits come as future service, not cash.** AWS says credits "will not entitle you to any refund or other payment."
- **You must claim.** Credits are not automatic, except AWS's rule that it will not charge for a single EC2 instance unavailable for more than six minutes of a clock hour.
- **Some public SLAs are only an effort statement.** Slack's page has no percentage or credit table, so ask what your plan's order form actually commits to.

## The key metrics to put in a vendor SLA

Pick only the metrics that match how the vendor can hurt your business.

| Metric | What it measures | How to define it | Best for |
|---|---|---|---|
| Uptime or availability | Share of time the service works | Minutes available divided by total minutes in a calendar month, excluding agreed maintenance; define "unavailable" (for Google, a user error rate above 5%) | SaaS, hosting, payment and phone systems |
| Response time | Time until a qualified person acknowledges an issue | From ticket creation to first human response, by severity | Any vendor with support |
| Resolution or workaround time | Time until the issue is fixed or a workaround is in place | From ticket creation to resolution or acceptable workaround, by severity | IT services, managed service providers, critical SaaS |
| Recovery point objective (RPO) | Maximum data you can afford to lose | NIST defines it as "the point in time to which data must be recovered after an outage" | Systems that hold transactions or records |
| Recovery time objective (RTO) | Maximum time to restore service after a disaster | How long systems can be in recovery before it hurts the business, per NIST SP 800-34 | Hosting, backup, ERP, core SaaS |
| Support hours and channels | When and how you can get help | Hours, time zone, holidays, phone or chat for severity 1 | All vendors |
| On-time delivery | Orders delivered by the agreed date | Share of orders delivered on or before the confirmed date | Goods suppliers, logistics |
| On-time in-full (OTIF) | Orders delivered on time, complete and without defects | Share of orders meeting all three conditions | Goods suppliers |
| Accuracy or quality | Error rate in delivered work | Defects, invoice errors or rework per period | Services, outsourcing, fulfillment |

Microsoft also shows that service levels need not be only uptime. Its Online Services SLA includes an email delivery time measure, with credits starting when average delivery time exceeds 1 minute.

### Severity levels: response is not resolution

Response time means someone is looking; resolution time means it is fixed. Ask for both, at least for the top two severities. The targets below are illustrative, not industry standards.

| Severity | Example definition | Illustrative response target | Illustrative resolution or workaround target | Hours covered |
|---|---|---|---|---|
| 1, critical | Service down or unusable for most users; no workaround | 1 hour | 4 hours | 24/7 |
| 2, high | Major function impaired; workaround is costly | 4 business hours | 1 business day | Business hours |
| 3, medium | Minor function impaired; workaround exists | 1 business day | 5 business days | Business hours |
| 4, low | Question or cosmetic issue | 2 business days | Next scheduled release | Business hours |

Define who decides severity. A fair rule is that the customer assigns it in good faith and the vendor can only lower it with your agreement.

## Downtime math: what each uptime level really allows

Uptime targets sound alike but allow very different amounts of downtime. The table uses an average month of 43,830 minutes (365.25 days divided by 12) and a 365-day year.

| Monthly uptime target | Allowed downtime per average month | Allowed downtime per 30-day month | Equivalent per year |
|---|---|---|---|
| 99% | About 438 minutes (7.3 hours) | 432 minutes (7.2 hours) | About 87.6 hours |
| 99.5% | About 219 minutes (3.7 hours) | 216 minutes (3.6 hours) | About 43.8 hours |
| 99.9% | About 43.8 minutes | 43.2 minutes | About 8.8 hours |
| 99.95% | About 21.9 minutes | 21.6 minutes | About 4.4 hours |
| 99.99% | About 4.4 minutes | 4.3 minutes | About 53 minutes |

**Method.** Allowed downtime equals total minutes in the period multiplied by (100% minus the target). Because most SLAs measure monthly, the yearly figure is only an equivalent: a vendor can use the full monthly allowance every month and still meet a monthly SLA.

Also check what counts as downtime. Microsoft excludes scheduled maintenance windows and "performance degradation or latency issues without actual service unavailability" unless a service has a performance target. A slow system that technically answers may count as up.

## Service credits, penalties and termination rights

Service credits are the usual consequence of a missed service level. Five terms decide whether they are worth anything.

### Credits as the sole remedy

Google, Microsoft and AWS all describe their credits as the sole and exclusive remedy for missing the SLA. That means a missed SLA alone generally will not let you claim your actual losses. If an outage could cost you far more than a month of fees, the answer is usually not a bigger credit but a termination right, a better architecture or a backup provider.

### Credit tiers

Tie larger credits to worse performance, as the published SLAs do, and add separate credits for missed severity 1 response or resolution targets.

### Caps

Most vendors cap credits. Microsoft caps credits at the monthly service fee for the service; Google caps them at 15 days of service per month. Ask that the cap applies per month, not per year, and that credits for different metrics can stack within it.

### Claim windows

Google requires a support case within 30 days; AWS needs the claim by the end of the second billing cycle, with request logs that "corroborate your claimed outage." Missing the window forfeits the credit. Ask for a longer window, or better, ask the vendor to apply credits automatically based on its own monitoring report.

### Termination for chronic failure

This is the most valuable clause a small customer can win. Define chronic failure in numbers, for example missing the uptime target in any three months in a rolling six, or two severity 1 incidents unresolved within target in a quarter. Then give yourself the right to terminate without an early termination fee and with a refund of prepaid, unused fees. If you plan to ask for fixed payments for misses beyond credits, have a lawyer review the wording, because how such payments are treated varies by jurisdiction. This guide is general information, not legal advice.

## Exclusions to read closely

Reasonable exclusions cover events outside the vendor's control, your own equipment or misuse, and beta features. Microsoft's list runs to 14 items, including third-party software and unsupported configurations. Push back on exclusions that remove the vendor's own failures:

- **Unlimited scheduled maintenance.** Cap it (for example, 4 hours a month), require advance notice and set windows outside your business hours. Microsoft states Exchange Online has no scheduled downtime.
- **Third-party providers the vendor chose.** If the vendor hosts on a cloud provider, that provider's outage is the vendor's risk, not yours.
- **Vague force majeure.** It should not cover failures a reasonable backup or disaster recovery plan would prevent.
- **Partial outages.** Make sure a broken core feature counts, not only a full outage.

## Reporting and review terms

Without reporting, the SLA depends on you noticing problems. Ask for:

- **A monthly service report**: uptime, incidents, tickets by severity with response and resolution times, and credits due.
- **Severity 1 incident reports** with root cause and prevention steps.
- **A status page** with incident history.
- **A quarterly review meeting** for critical vendors, plus an annual review of the targets themselves.

Those reports become the raw data for a [vendor scorecard](https://businesslang.com/guides/how-to-build-a-vendor-scorecard/), which turns SLA results into a regular performance review.

## Step 1: Rank the services you depend on

List what the vendor provides and which parts would stop revenue, payroll, shipping or customer service if they failed. Only those parts need strict targets. Our [vendor risk management guide](https://businesslang.com/guides/vendor-risk-management-b2b-purchasing/) explains how to tier vendors by impact.

## Step 2: Collect the vendor's standard SLA and compare it

Ask for the current SLA, its version date and the plan or tier it applies to. Compare it with the published SLAs above. If you are still choosing a vendor, put your required service levels into the requirements section of your [RFP](https://businesslang.com/guides/how-to-write-an-rfp/) so bidders price them in.

## Step 3: Define each metric precisely

For each metric, write the definition, the measurement method, the measuring tool, the period (calendar month is common) and who owns the data. Define "unavailable" in a way your users would recognize.

## Step 4: Set targets you can justify

Translate each uptime target into minutes using the table above, and match response and resolution times to how long your business can tolerate the problem. Set RPO and RTO from your own recovery needs, then check the vendor's backup and recovery approach supports them.

## Step 5: Attach consequences

Add a credit table, the cap, the claim process and window, and a chronic failure termination right with a refund of prepaid fees.

## Step 6: Narrow exclusions and add reporting

Cap maintenance, remove exclusions covering the vendor's own subcontractors, require monthly and incident reports, and state that service levels cannot be reduced during the term without your written agreement. Record which dated SLA version your contract uses.

## Step 7: Negotiate, then diarise

Prioritize the asks that matter most: termination for chronic failure, longer claim windows and automatic credits usually come before higher uptime. Our guide to [negotiating SaaS pricing and renewals](https://businesslang.com/guides/how-to-negotiate-saas-pricing-and-renewals/) covers timing and leverage. Once signed, add claim deadlines and review dates to a shared calendar.

## Sample SLA clause language (illustrative)

The clauses below are illustrative starting points, not legal advice. Adapt them with counsel.

**Availability.** "Vendor will make the Service available at least 99.9% of the time in each calendar month, measured as total minutes in the month minus minutes of Downtime, divided by total minutes in the month. Downtime means any period in which Customer's users cannot log in to or complete core transactions in the Service, as shown by Vendor's monitoring or reasonably documented by Customer."

**Scheduled maintenance.** "Scheduled maintenance will not exceed four hours per calendar month, will occur between 10 p.m. and 5 a.m. Customer local time, and requires at least five business days' written notice. Maintenance outside these limits counts as Downtime."

**Support.** "Vendor will respond to Severity 1 incidents within one hour, 24 hours a day, 7 days a week, and will provide a resolution or acceptable workaround within four hours."

**Service credits.** "If Vendor misses the availability target, Vendor will credit Customer 10% of the monthly fees for the affected Service for availability below 99.9%, 25% below 99.5% and 50% below 99.0%. Vendor will apply credits automatically to the next invoice based on its monthly service report. Credits will not exceed 50% of the monthly fees for that month."

**Chronic failure.** "If Vendor misses the availability target in any three months in a rolling six-month period, Customer may terminate the Agreement on written notice without penalty, and Vendor will refund any prepaid fees for the period after termination."

**Reporting and review.** "Vendor will deliver a service report within ten business days after each month end and a root cause report within five business days after any Severity 1 incident. The parties will review performance against this SLA quarterly."

## Illustrative example: what a 3-hour outage is worth

**Method.** Illustrative numbers only. A business pays $2,000 a month for an operations SaaS tool. In a 30-day month (43,200 minutes), the tool is down for 3 hours (180 minutes), so uptime is (43,200 minus 180) divided by 43,200, or about 99.58%.

- **Microsoft-style tiers** (25% below 99.9%): a credit of $500.
- **Google-style tiers** (3 days of service below 99.9%): roughly 3/30 of the monthly fee, or about $200.
- **AWS Region-Level-style tiers** (10% below 99.99%): $200.
- **The sample clause above** (10% below 99.9%; the 25% tier starts below 99.5%): $200.

If those 3 hours stopped $15,000 of orders, no credit comes close. That is why the termination right, the claim process and a workable recovery plan matter more than the exact credit percentage.

## The bottom line

A clear vendor SLA is short and specific: the services that matter, measurable targets with definitions, a narrow set of exclusions, credits you can actually claim, a right to leave after repeated failure, and regular reports reviewed on a schedule. Since 99.9% uptime with credits as the sole remedy is the norm, spend your negotiating effort on claim windows, maintenance limits and chronic failure termination. Then track results over time with a [vendor scorecard](https://businesslang.com/guides/how-to-build-a-vendor-scorecard/) and bring the evidence to every renewal.

## Frequently asked questions

**Q: What is the difference between an SLA and a contract?**

A: The contract (often a master services agreement plus an order form) sets the overall legal relationship: price, term, liability, data protection and termination. The SLA is a schedule or policy within that relationship that defines measurable service levels and what you get if they are missed. Check that the contract actually incorporates the SLA and says which document wins in a conflict.

**Q: What is a good uptime percentage for a SaaS SLA?**

A: For business SaaS, 99.9% monthly uptime is a common published commitment. Google Workspace and Microsoft 365 services such as Exchange Online, Teams and SharePoint Online all use 99.9% (checked November 13, 2026). Ask for more only where an outage would stop revenue, and expect to pay for it.

**Q: What are service credits in an SLA?**

A: Service credits are a reduction in future fees when the vendor misses a service level. Microsoft credits 25%, 50% or 100% of the monthly fee for its Online Services depending on how far uptime falls, while Google Workspace adds 3, 7 or 15 days of service. Credits are usually applied to future invoices, not paid as cash.

**Q: Can small businesses negotiate an SLA?**

A: Standard SLAs from very large vendors are rarely changed for small customers. Smaller and mid-sized vendors, managed service providers, agencies and logistics or supply partners are often more flexible, especially on support response times, reporting, claim windows and termination for repeated failures.

**Q: What should happen if a vendor keeps missing its SLA?**

A: A good SLA defines chronic failure, for example missing the same target in three months out of any six, and gives you the right to terminate without penalty and get a refund of prepaid fees for the remaining term. Without that clause, credits may be your only remedy.


## Sources

1. [Google Workspace Service Level Agreement, last modified August 31, 2026 (checked November 13, 2026)](https://workspace.google.com/terms/sla.html)
2. [Microsoft: Service Level Agreements (SLA) for Online Services, October 2026 consolidated SLA (checked November 13, 2026)](https://www.microsoft.com/licensing/docs/view/Service-Level-Agreements-SLA-for-Online-Services)
3. [AWS: Amazon Compute Service Level Agreement, last updated May 25, 2022 (checked November 13, 2026)](https://aws.amazon.com/compute/sla/)
4. [Slack: Service Level Agreement, effective July 11, 2024 (checked November 13, 2026)](https://slack.com/terms/service-level-agreement)
5. [NIST CSRC Glossary: Recovery Point Objective, citing NIST SP 800-34 Rev. 1 (checked November 13, 2026)](https://csrc.nist.gov/glossary/term/recovery_point_objective)
6. [NIST CSRC Glossary: Recovery Time Objective, citing NIST SP 800-34 Rev. 1 (checked November 13, 2026)](https://csrc.nist.gov/glossary/term/recovery_time_objective)
