How to Build an SLA Report: Formulas and Template
Key takeaways
- SLA attainment is uptime minutes divided by total minutes in the period; a 30-day month has 43,200 minutes, so a 99.9% target allows 43.2 minutes of downtime.
- Report attainment and breach count separately, because total minutes lost and the number of distinct incidents that crossed your SLA threshold are different numbers.
- Compute attainment per component and weight by billing or usage share instead of averaging SLAs across services.
- Attribute every breach to one of four buckets (your app, your vendor, a third party, or force majeure), and attribute ambiguous breaches against yourself.
- Write exclusion rules for scheduled maintenance into the SLA text before the first reporting period, and log windows as timestamped events in your monitoring tool.
An SLA report tells your customers whether you met the uptime and response-time commitments in your contract, using numbers you can defend: SLA attainment, breach counts, and downtime minutes per component. This guide gives you the exact formulas, breach attribution rules, and a copy-paste template with worked examples, so you can generate the report from your monitoring stack every month instead of rebuilding it.
What is an SLA report and who is it for?
An SLA report is a periodic record of the uptime you actually delivered against the uptime you promised: availability percentage, response time, incident counts, and any breaches with their cause. It turns "we had a rough month" into a number both sides can check.
Customer-facing vs internal SLA reports
The customer-facing version shows attainment, downtime minutes, and breach attribution in plain terms, often as a PDF attached to an invoice. The internal version adds error budget burn, false alarm rates, and which probes caught the outage versus which missed it. Same data, different framing.
When a 99.9% promise becomes a contract term
AWS defines 99.9% as 43 minutes and 12 seconds of monthly downtime (AWS SLA documentation). Write that number into a contract and you need the report ready before a customer asks, not during a dispute. If your monitoring is still homepage-only URL checks, protocol-level probes with scheduled PDF reports give you the raw data to back the claim.
Which metrics belong in an SLA report?
An SLA report needs five metric families: uptime percentage, response time percentiles, breach counts, MTTR, and error budget consumption. Customers care about the first three; your team needs all five to actually fix things. Anything else is padding.
Uptime and availability
Uptime percentage is the headline number: successful probes divided by total probes over the reporting period. Report it per service, not just as one blended figure, because your API can be down while your dashboard is fine. Measure it against the SLA you actually promised, not a round 99.99% you never committed to.
Response time percentiles (p95, p99)
Averages hide your worst minutes, and your worst minutes are what customers notice. Use p95 for customer reports and p99 internally. Set the threshold at the slowest response your product can have before users feel it, then alert on breaches of that threshold.
Breach counts and MTTR
Count every SLA breach, when it started, and how long it took to resolve. MTTR (mean time to resolve) tells customers how quickly you recover, which matters more to them than how rarely you fail. A report showing four short incidents often reads better than one showing a single six-hour outage.
Error budget burn
If your SLA is 99.9%, your error budget is 43.8 minutes of downtime per month. Track how much of it you have spent and how fast. Internal reports should flag when burn rate exceeds schedule, because that is your early warning before the next breach.
Customer vs internal: what goes where
| Metric | Customer report | Internal report |
|---|---|---|
| Uptime % per service | Yes | Yes |
| p95 response time | Yes | Yes |
| p99 response time | No | Yes |
| Breach count and duration | Yes | Yes |
| MTTR | Yes | Yes |
| Error budget burn | No | Yes |
| False alarm rate | No | Yes |
False alarms belong only in the internal report. If your probes fire on non-issues, your MTTR and breach counts inflate, and you will either apologize for nothing or stop trusting your own alerts. StatusPulse generates both report types from the same probe data, with white-labelled versions for client-facing delivery (https://statuspulse.ai/features/reports-whitelabel).
How do you calculate SLA attainment and breach rate?
Basic attainment formula with a worked example
SLA attainment is uptime minutes divided by total minutes in the reporting period. A 30-day month has 43,200 minutes (30 × 24 × 60, per timeanddate.com's calendar reference). If your API was down 47 minutes in September, attainment is (43,200 − 47) / 43,200 = 99.89%. To hit 99.9% you can lose 43.2 minutes per month; to hit 99.95%, only 21.6. That asymmetry is why breach attribution matters: you need to know whether those 47 minutes were your code, your host, or a third-party DNS provider before you credit or eat them.
Weighted component-level SLAs
If you promise 99.9% on the API and 99.5% on the dashboard, don't average them. Compute attainment per component, then weight by each component's share of the customer's bill or usage. Example: API at 99.92% weighted 80%, dashboard at 99.30% weighted 20% gives (0.9992 × 0.8) + (0.993 × 0.2) = 99.80%. The dashboard miss drags the blended number down even though the headline "uptime" looked fine. StatusPulse's scheduled reports break attainment down per probe so this math is automatic rather than a spreadsheet exercise.
Breach rate vs attainment: they are not the same number
Attainment measures total minutes lost; breach rate counts how many distinct incidents crossed your SLA threshold. You can post 99.99% attainment and still breach twice if two short outages each exceeded the allowed window, or miss attainment entirely from one long outage that never breached a per-incident cap. Report both: attainment drives service credits, breach count drives the conversation about reliability trends.
How should you structure an SLA report for customers vs internal teams?
The 7-section report template
Use the same skeleton for both audiences:
- Summary header — reporting period, service name, SLA target, and the headline attainment figure (e.g. 99.94% vs 99.9% committed).
- Attainment table — per-service uptime, average and p95 response time, breach count, and total breach minutes.
- Incident log — timestamped list of every incident: start, detection, mitigation, resolution, customer-impact minutes.
- Breach attribution — for each breach, the cause column: your fault, vendor fault, customer-side, or scheduled maintenance.
- Response-time trend — a simple chart so customers see degradation before it becomes an outage.
- Exclusions and credits — maintenance windows excluded, and any SLA credits issued.
- Appendix — probe locations, check intervals, and methodology so the numbers are reproducible.
What to keep and cut for the customer version
Keep sections 1, 2, 3, and 6. Customers care about attainment, what broke, when, and what you owe them. Cut the appendix and raw probe data; internal teams need those to debug, customers don't. Keep breach attribution but soften the wording, not the facts: "third-party DNS provider failure" is fine, a blame essay is not. If you generate these monthly, StatusPulse's report builder with AI summary drafts the narrative section from your incident log, which saves the most time (https://statuspulse.ai/features/reports-ai-summary).
Breach attribution: writing the cause column honestly
The cause column is where SLA reports get gamed, and customers notice. Attribute every breach to exactly one of four buckets: your application, your infrastructure vendor, a third-party dependency, or force majeure. If a breach is ambiguous, attribute it against yourself and say so. Under-claiming attainment by 0.02% costs less in credits than one disputed report costs in trust.
What reporting period and granularity should you choose?
Monthly vs quarterly vs rolling windows
Calendar months are the default because customers understand them and invoices line up, but a bad 48 hours can wipe out your entire error budget in one period. Rolling 30-day windows smooth this out and are what most uptime widgets display, though they make breach attribution harder: an incident can straddle two reports. Quarterly reports suit enterprise customers with contractual penalties, but 1-10 person teams rarely have the incident volume to make them useful. Pick monthly for customer-facing SLA reports and keep a rolling window on your internal dashboard so a single outage never hides behind a fresh period.
Probe frequency and granularity pitfalls
A 60-second check interval means a 59-second outage can go unnoticed, and per-probe granularity overstates downtime when one region fails but customers are fine. If you publish an SLA, calculate attainment from your shortest probe interval and note the caveat in the report. Google's SRE book frames this as an error budget: the downtime you're allowed to spend per period, measured honestly (Error Budget chapter). StatusPulse reports at per-probe granularity so you can show which probe flagged the breach, and its status pages render in your customers' language, not just English, via multi-language support.
How do you handle exclusions, maintenance windows, and disputed breaches?
An SLA report that counts every minute of downtime, including downtime you scheduled or didn't cause, will understate your attainment and invite arguments. Define exclusion rules in writing before the first reporting period.
Scheduled maintenance: exclude or count?
Pick one rule and state it in the SLA text. Most B2B SaaS teams exclude maintenance announced at least 48 hours in advance, capped at a stated number of hours per month. Anything announced late counts as downtime. Log maintenance windows as events in your monitoring tool, not in a spreadsheet, so the exclusion is timestamped and auditable.
Customer-caused and third-party downtime
Attribute each breach: your infrastructure, the customer's own misconfiguration (bad DNS, expired card, disabled webhooks), or a named third party. Only the first category counts against your SLA. A protocol-level probe can tell you whether your endpoint answered; it cannot tell you why a customer's integration failed, so keep the raw probe data as evidence either way.
Documenting disputed breaches
For every breach, record start time, end time, affected endpoints, probe evidence, and root cause within 24 hours. If a customer disputes, you negotiate from timestamps, not memory. Atlassian Statuspage's own plans (pricing) charge up to $199/mo and still leave incident evidence to you, so build this log regardless of vendor.
How do you automate SLA reporting from your monitoring stack?
Manual spreadsheet approach
Export incident timestamps from your monitoring tool, paste into a sheet, and compute attainment with the same formulas from earlier sections. It works at one customer and one month, but it breaks the moment you have several SLAs, maintenance windows, and disputed breaches. Budget roughly an hour per report per customer.
Built-in reporting vs DIY
UptimeRobot and Better Stack can export or display uptime history, but you will still assemble the SLA math yourself. StatusPulse generates uptime and incident data per probe, yet it does not produce a finished, customer-facing SLA report today. The realistic middle path is a scheduled script that pulls incident logs via API into a templated doc, so the only manual step is reviewing breach attribution before it ships.
Honest limits of automated reports
Automation cannot decide whether a breach was your fault, the customer's, or a third party's. It also cannot exclude a maintenance window it was never told about. Treat automated reports as a draft: the numbers are mechanical, but every exclusion and disputed incident needs a human pass before a customer sees it.
FAQ
How many 9s is 99.9% uptime in minutes per month?
99.9% means you can lose 43.2 minutes per 30-day month, since a 30-day month has 43,200 minutes and 0.1% of that is 43.2. Each additional 9 divides the budget by ten: 99.95% allows 21.6 minutes, 99.99% allows only 4.32. That asymmetry is why breach attribution matters at higher targets.
Do scheduled maintenance windows count against my SLA uptime?
Only if you let them. Pick one rule and write it into the SLA text: most B2B SaaS teams exclude maintenance announced at least 48 hours in advance and capped at a stated number of hours per month. Anything announced late counts as downtime. Log the windows as timestamped events in your monitoring tool so the exclusion is auditable.
Can I offer a 99.9% SLA if my monitoring only checks every 5 minutes?
You can, but you will undercount downtime. A 5-minute check interval can miss outages shorter than 5 minutes entirely, so your measured attainment will overstate the real figure. If you publish an SLA, calculate attainment from your shortest probe interval and note the check interval in the report's methodology appendix.
What should I do when a customer disputes a breach my report shows as passing?
Negotiate from timestamps, not memory. Keep a log for every incident with start time, end time, affected endpoints, probe evidence, and root cause recorded within 24 hours. If the breach is ambiguous, attribute it against yourself and say so; under-claiming attainment by a few hundredths of a percent costs less in credits than one disputed report costs in trust.