SLA Compliance for SaaS: Measure It and Report It
Key takeaways
- A 99.9% monthly SLA allows exactly 43.2 minutes of downtime in a 30-day month, so one 47-minute outage is a breach.
- Rewrite vague contract promises as specific metrics: what you probe, how often, what counts as failure, and over what window.
- Track uptime per component and per contract, because a single global uptime number can hide a per-customer breach.
- Use 60-second probes and keep 90 days of raw CSV data so disputed SLA claims are settled with evidence, not opinions.
- Freeze non-critical deploys when your remaining monthly downtime budget drops below one incident's worth.
SLA compliance means proving, with data, that your uptime met the number written in your customer contracts, usually 99.9% monthly. This guide shows you how to turn that contract language into measurable metrics, run the error budget math for a real outage, and track compliance per customer without enterprise tooling.
What Does SLA Compliance Actually Mean for a SaaS?
An SLA is a contractual promise: a specific uptime percentage, measured over a defined period, with penalties when you miss it. Compliance means your measured availability met that number and you can prove it with data the customer accepts. It differs from an SLO, which is an internal target you set for yourself, and from the "99.9% uptime" badge on your homepage, which is marketing with no measurement methodology behind it. The Google SRE book defines the distinction well: SLAs are external commitments with consequences, SLOs are internal targets, and the gap between them is your error budget.
SLA vs SLO vs uptime percentage: the terms customers will hold you to
Uptime percentage alone is ambiguous until you define the measurement window, the probe method, and what counts as downtime. A customer with a signed 99.5% SLA can dispute any number you report unless your methodology is documented. Treat the SLO as your safety margin: set it above the contractual SLA so ordinary bad weeks don't trigger breach clauses.
What a breach triggers: credits, termination clauses, and refunds
Most small-team SLAs offer service credits, typically a percentage of the monthly fee, as the remedy. Some contracts add termination rights after repeated breaches, which is why tracking compliance per customer matters before legal does it for you. For a practical walkthrough of turning these clauses into tracked metrics, see SLA compliance monitoring: a practical guide to measuring service commitments.
How Do You Turn SLA Contract Language into Measurable Metrics?
"Highly available" is not a metric. "99.9% of successful HTTP responses measured over a rolling 30 days" is. Rewrite every commitment as: what you probe, how often, what counts as failure, and over what window. A vague SLA is unenforceable in both directions, so precision protects you as much as the customer.
Pick the measurement window: rolling 30 days vs calendar month
A rolling 30-day window smooths out one bad day; a calendar month gives customers a fixed reporting period. Calendar months are easier to invoice credits against, which is why most small SaaS teams use them. Pick one and state it in the contract.
Define what counts as downtime: HTTP status, latency thresholds, and protocol-level failures
| Failure type | Example | Counts as downtime? |
|---|---|---|
| HTTP status | 5xx responses | Yes |
| Latency threshold | API responses over 2s | Yes, if you write it in |
| Protocol-level | TCP connect fails, TLS handshake errors | Yes |
| Content check | Login page returns 200 but renders an error | Yes, if probed |
URL checks miss the last two rows. Protocol-level probes catch a server that accepts connections but never serves a valid response, which is the failure mode that generates 3 AM incidents.
Write the exclusion list: scheduled maintenance, customer-caused failures, force majeure
Excluded minutes come off the denominator of your uptime calculation, so define them narrowly: maintenance with 48 hours notice, failures caused by the customer's own configuration, and named force majeure events. Anything not listed is your downtime.
For the tooling that produces this data, see our guide to SLA monitoring tools and compliance.
The Error Budget Math: A Worked Example
A 99.9% SLA does not mean "almost never down." It means you are contractually allowed 43.2 minutes of downtime in a 30-day month (43,200 minutes × 0.1%). At 99%, you get 432 minutes; at 99.5%, 216 minutes; at 99.95%, only 21.6 minutes. Every tier you advertise is a promise about a specific number of minutes, so write that number down before a customer writes it into a renewal negotiation.
Monthly downtime allowances at 99%, 99.5%, 99.9%, and 99.95%
Using a 30-day month: 99% allows 7.2 hours, 99.5% allows 3.6 hours, 99.9% allows 43.2 minutes, and 99.95% allows 21.6 minutes. Pick the tier you can actually defend with probe data, not the one that sounds good on a pricing page next to Atlassian Statuspage's $29-199/mo plans.
A worked month: one 47-minute outage against a 43.2-minute budget
Say you commit to 99.9%. On the 12th, a database failover takes your API down for 47 minutes. That single incident burns 47 of your 43.2 allowed minutes, putting you at 99.892% for the month before anything else goes wrong. One incident, one SLA breach, and 18 days still left on the calendar. This is why a single unmeasured deployment hiccup matters: at this tier you have no room for a second bad afternoon, and a 10-minute blip on the 20th pushes you to 99.87% with no recovery mechanism except next month's fresh budget.
What happens when the budget goes negative
Once you cross the line, the math stops being about uptime and starts being about credits and trust. Typical SaaS SLAs pay 10-30% of the monthly fee as a service credit per breach tier (illustrative; check your own contract language). The operational fix is to treat the budget as a deploy throttle: when remaining downtime for the month drops below one incident's worth, freeze non-critical releases. Track the burn rate per customer, not just fleet-wide, because the customer who saw the 47-minute outage is the one who files the claim, and your status page data is your only defense.
How Do You Track SLA Compliance Per Customer?
Your status page shows one global uptime number, but your SLA is signed with individual customers. If a premium customer's API region degrades while the rest of your service runs fine, the global number says 99.98% while that customer experienced a breach. Per-customer tracking starts with what you probe, not what you report.
Component-level and region-level probes as the building blocks
Probe each customer-facing component separately: API, dashboard, webhook delivery, background jobs. Add region-level probes if you deploy per-region, so a us-east outage doesn't get averaged away by healthy EU traffic. Protocol-level checks (TCP, DNS, SMTP) matter here too, because a URL check on your homepage can stay green while a customer's SMTP integration is down.
Mapping customer contracts to the components they depend on
Keep a simple mapping: contract → tier → components. A customer paying for 99.9% on the API and webhooks has an SLA computed only from those two probes' uptime, not from your global average. A spreadsheet works at small scale; the mapping just has to exist and survive an incident.
Why a single global uptime number can hide a per-customer breach
Global uptime is an average across components and regions, and averages hide tails. Two components at 99% and 100% both report "99.5% overall," but only one customer base was breached. Track uptime per component, then aggregate per contract, and your status page becomes evidence rather than a summary. The incident side of this is covered in the 8-phase incident management lifecycle.
What Monitoring Setup Produces Defensible SLA Data?
Probe interval vs SLA granularity: why 5-minute checks can't prove a 99.9% commitment
A 99.9% monthly SLA allows 43.2 minutes of downtime. With 5-minute checks, a single missed interval can hide up to 4 minutes 59 seconds of outage, and your error budget math becomes an estimate, not evidence. Use 60-second probes (or 30 seconds on your Pro plan) so your reported downtime stays within seconds of what the customer experienced. Anything coarser and a disputed invoice turns into your word against theirs.
Protocol-level checks beyond HTTP: TCP, DNS, database, and heartbeat probes
A homepage returning 200 says nothing about your API, Postgres, or mail relay. Monitor the protocols your SLA actually names: TCP checks for port-level services, DNS resolution checks, database connection probes, and heartbeats for background jobs. UptimeRobot and Better Stack both cover HTTP plus some protocol checks (Better Stack pricing starts at $25-85/mo); StatusPulse includes protocol-level probes from $5/mo. Honest limitation: StatusPulse has no browser-level transaction monitoring, so multi-step checkout flows need synthetic checks elsewhere.
Data retention and export: keeping raw incident evidence
Keep raw probe results for at least 90 days and export them as CSV. When a customer disputes an SLA credit, you hand over timestamps, response codes, and probe locations, not a summary chart. Pair this with documented maintenance windows so scheduled work never counts against the budget.
How Do You Wire SLA Reporting into Your Incident Workflow?
Incident start and end timestamps as the source of truth
Your SLA clock starts when the probe fails, not when someone opens a Slack thread. StatusPulse records detection time automatically, and the responder marks resolution in the incident UI. If you let engineers edit timestamps after the fact, your SLA data becomes an opinion. Keep the probe-detected start immutable and log any human-confirmed start separately.
Scheduled maintenance windows and how they interact with the SLA clock
Decide once whether maintenance counts against uptime, and write that into the contract language. Most SaaS SLAs exclude announced windows, so StatusPulse subtracts scheduled maintenance from the denominator automatically when it overlaps an incident. The catch: if you don't announce the window on your status page beforehand, treat it as an outage. Auditors and customers check this.
Monthly report contents: uptime, incidents, MTTR, and credits issued
A defensible report has five lines: measured uptime per probe, incident list with start/end/cause, MTTR, maintenance minutes excluded, and any service credits owed. If a customer disputes the number, the incident timeline is your evidence. Tools like UptimeRobot and Hyperping publish monitoring plans from free tiers upward (pricing comparison), but few generate per-customer SLA reports, so check before you switch.
How Do Small Teams Run SLA Compliance Without Enterprise Tooling?
Honest comparison: StatusPulse, Statuspage, Better Stack, UptimeRobot, Hyperping, and Instatus against the same criteria
The criteria from earlier sections: protocol-level probes, defensible uptime data, per-incident reporting, and a price a solo founder can carry. Atlassian Statuspage and Better Stack cover the reporting well but start at $29-199/mo and $25-85/mo respectively. UptimeRobot is cheap monitoring but thin on status pages. Hyperping does both. Instatus leans status-page-first. StatusPulse covers monitoring plus status pages from $5/mo, with protocol-level probes on paid tiers.
Where StatusPulse falls short for SLA work
Honest limits: there is no built-in per-customer SLA ledger, so you export probe data and do the error budget math in a spreadsheet. PagerDuty-style escalation is thinner than Better Stack's on-call tooling. If you need contractual audit trails or multi-region probe attestations, an enterprise platform still wins.
A minimum viable SLA stack for a solo founder
Run 5 free probes on your critical paths (API, auth, database). Upgrade to Starter at $5/mo for the public status page. Export monthly uptime, compute SLA per customer in a sheet, and paste the summary into incident postmortems. That is SLA compliance a 1-10 person team can actually sustain.
FAQ
How is SLA compliance calculated when an outage spans two months?
Count each minute against the month it occurred in. A 30-minute outage crossing midnight on the 1st puts 20 minutes in one month's budget and 10 in the next. With calendar-month windows, each month gets its own allowance, 43.2 minutes at 99.9%, so a single incident can breach two reporting periods.
Do scheduled maintenance windows count against my SLA uptime?
Only if your contract says they do. Most small SaaS SLAs exclude announced windows, which removes those minutes from the denominator of your uptime calculation. The catch: announce the window on your status page beforehand with the agreed notice period, typically 48 hours. An unannounced window is treated as an outage by customers and auditors alike.
What happens if I miss an SLA and can't afford service credits?
Credits are usually 10-30% of the monthly fee per breach tier, so the exposure is bounded, not catastrophic. The real cost is termination rights after repeated breaches. Freeze non-critical deploys when remaining downtime drops below one incident's worth, and negotiate credit caps into the contract before renewal, not after a breach.
Can I use my status page uptime number as proof of SLA compliance in a customer dispute?
A single global uptime number is weak evidence because averages hide per-component and per-region breaches. What holds up in a dispute is raw probe data: timestamps, response codes, and probe locations exported as CSV, ideally covering 90 days or more. Pair it with the incident timeline and documented maintenance exclusions, then compute uptime per contract.