Incident Uptime Band Reporting: A Guide for Small SaaS Teams
Key takeaways
- Incident uptime band reporting splits uptime into full outage, partial degradation, and scheduled maintenance bands instead of one blended figure.
- Weight partial-degradation minutes by the affected fraction you published during the incident, not a number reconstructed afterward.
- Overlapping incidents count once in the most severe band, with each band's denominator staying the full reporting window.
- Only maintenance announced 48 hours ahead, customer-caused failures, and contractually named force majeure can be excluded, and every exclusion needs a documented row.
- Make the blended number the contractual SLA metric and the band breakdown the diagnostic layer reported alongside it.
Incident uptime band reporting means splitting your uptime number into severity-based bands, like full outage, partial degradation, and scheduled maintenance, instead of publishing one blended figure. This guide shows how to define the bands, compute uptime per band from raw incident logs, and handle partial outages, overlapping incidents, and documented exclusions with a fully worked example.
What is incident uptime band reporting and how does it differ from a single blended uptime number?
Incident uptime band reporting splits your uptime into severity bands: full outage, partial degradation, and scheduled maintenance. Each band gets its own uptime figure, computed from raw incident data, instead of one blended number that averages everything together.
Why a 99.95% blended number can still mean a bad month for one customer segment
A blended 99.95% allows about 21 minutes of downtime per month, but it says nothing about who was affected. If your API went fully down for 15 minutes while the marketing site stayed up, the blend hides a severe incident behind a passing grade. Customers on the API saw an outage; your dashboard claims you barely slipped.
Band reporting fixes this by scoring each incident by impact. A full outage counts against the outage band; a slow endpoint counts against partial degradation. That distinction matters when your status page and monitoring come as a package: Atlassian Statuspage charges $29–199/month (pricing) for a single operational banner, while cheaper tools like StatusPulse give you the incident history to compute bands yourself. We cover the full method on the StatusPulse blog.
How do you define the bands (full outage, partial degradation, scheduled maintenance) for your service?
Map each band to a component impact model, not a feeling. Start by listing your components (API, dashboard, database, background jobs), then define bands by what a customer can actually do during the incident:
| Band | Definition | Example |
|---|---|---|
| Full outage | Core path unusable for all customers | API returns 5xx for every request |
| Partial degradation | Core path works, a component is degraded or a subset of customers affected | Webhooks delayed, or one region slow |
| Scheduled maintenance | Announced work with a start time and expected duration | Database migration, Sunday 02:00–02:30 UTC |
Tie bands to components, not vibes. If an incident touches two components, it belongs to the more severe band, and severity maps to impact: full outage = SEV1, partial degradation = SEV2. Write the mapping down before your next 3 AM incident, because nobody classifies calmly mid-outage.
Where scheduled maintenance sits: it is a band you report separately, never blended into incident uptime. Announce it on your status page, and if you publish SLA reports, exclude maintenance windows explicitly with timestamps. StatusPulse lets you export incident data and generate per-band PDF uptime and SLA reports, so the band definitions in your docs match what customers receive.
How do you calculate uptime per band when incidents overlap or affect only some customers?
Worked example: one month (43,200 minutes), four incidents, three bands.
The raw incident log
- Oct 7, 45 min: API down for all customers → full outage
- Oct 14, 120 min: job queue degraded, 1 of 4 components → partial
- Oct 21, 40 min: API errors for one region, 25% of customers → partial
- Oct 28, 50 min: API down; a dashboard partial (03:40–03:50) runs inside it
Band 1: full outage
45 + 50 = 95 minutes fully down. 95 / 43,200 puts the full-outage band at 99.78%. Report that number per band, not blended into one figure.
Band 2: partial degradation weighted by component impact
Partials are where blended uptime lies, so weight each one by impact. Weighted minutes = wall-clock minutes × affected fraction: the Oct 14 job-queue degradation counts as 120 × 25% = 30 weighted minutes, and the Oct 21 regional errors count as 40 × 25% = 10. The fraction you use is the one you published on the status page during the incident, not a number reconstructed afterward; customers judge you against what you told them at 3 AM, and re-deriving impact after the fact is how trust dies. Record the fraction in the incident record when you open it, then export the log as CSV so every weighted minute is auditable (incident CSV export). Band 2 uptime = (43,200 − 40) / 43,200 = 99.91%.
Overlapping incidents: count the wall-clock window once
The Oct 28 full outage contains a 10-minute dashboard partial; those minutes count in the full-outage band only, because customers experienced the worse state. Each band's denominator stays the full month, so subtract each band's windows once, never twice. StatusPulse tiers run free (5 probes, 1 page), $5/mo Starter, $19/mo Pro (pricing).
Which windows can you legitimately exclude from a band, and how do you document exclusions?
Three categories qualify: maintenance announced at least 48 hours ahead, downtime caused by the customer's own misconfiguration or credential failure, and force majeure events your contract names explicitly. Anything else stays in the band. If an incident overlaps a maintenance window, the incident counts; the maintenance exclusion covers only the window you announced, not the outage you caused during it.
The exclusion log format
Keep one row per exclusion: timestamp range in UTC, band affected, category, the announcement or contract clause that authorizes it, and who approved it. A spreadsheet works at your size. The test is whether a customer reading your SLA report six months later can trace every excluded minute to a documented reason without emailing you.
Why silent exclusions destroy trust
Excluding minutes without publishing them turns your 99.95% claim into a number customers can't verify, and they will assume the worst when a real outage contradicts it. Publish exclusions alongside each band's uptime figure, and set alert thresholds that account for excluded windows so false alarms don't push you into sloppy, undocumented exclusions at 3 AM (automating that threshold logic helps here).
How do you present band-based uptime to customers and stakeholders without eroding trust?
Showing three bands instead of one number can look like you're hiding something, so the format matters. Lead with the worst band and explain the math in one sentence.
The monthly band report format
Publish a table with one row per band: full outage minutes, partial degradation minutes, scheduled maintenance minutes, and uptime per band against target. Add an error-budget line for each band so customers see how much slack remains, not just what you missed. If a partial outage hit only 20% of customers, say so; a blended 99.95% that hides a 40-minute full outage reads worse than an honest 99.9%.
Public page vs contract report
Keep the public page qualitative: incident history, band definitions, and current status. Send contract reports monthly with raw minutes, exclusions, and credits. Pricing context helps here: Atlassian Statuspage runs $29-199/mo and Better Stack charges $25-85/mo for monitoring and status pages (Better Stack pricing), so a $5/mo StatusPulse Starter plan with the same transparency is defensible. Pair it with SMS alerts so stakeholders hear about the outage before the report.
How do you automate band reporting from incident data in your monitoring stack?
Most monitoring tools already record what you need; the work is exporting it and classifying each incident into a band.
What your incident data must capture
For band reporting to work, your status page or monitoring tool must record per incident: start time, end time, severity (full outage vs. partial degradation), affected components, and whether it was scheduled maintenance. Tools like Atlassian Statuspage and Better Stack export incident history; check that your export includes resolved timestamps and component impact, not just titles. If your current tool only logs "down at 3:12, up at 3:40" with no partial-outage flag, you cannot compute bands retroactively.
A minimal spreadsheet/script pipeline
Export incidents to CSV, then run a short script (Python or even spreadsheet formulas) that: assigns each incident a band from its severity field, computes affected minutes per band, subtracts documented exclusions by ID, and divides by total window minutes. A 50-line script handles a year of incidents for a small team. Refresh it monthly or on demand; the pipeline is simple enough to rebuild from scratch if your tool changes exports.
StatusPulse stores severity and component impact per incident, so the CSV maps directly to bands with no manual tagging.
When does band reporting conflict with contractual SLA language, and how do you reconcile the two?
Most contracts you signed or inherited specify a single uptime percentage, often measured monthly. Band reporting splits that number by severity, so a 99.9% month could show 99.95% for full outages and 99.7% for partial degradation. That is not a violation, but it is a mismatch: your contract's remedy clause (credits, termination rights) keys off the blended figure, not the bands.
The fix is to make the blended number the contractual metric and the bands the diagnostic layer. Report both: one number that satisfies the contract, plus the band breakdown that explains it.
Amendment language that works
Add a short clause stating that uptime is measured per the published severity definitions, that the contractual percentage is the blended total, and that band-level figures are informational. Keep the severity definitions in a versioned public document, not the contract itself, so you can refine probes without renegotiating. If you are drafting from scratch, base definitions on a recognised standard rather than inventing thresholds, and link the definitions from your status page so customers can see the same data you report.
Which tools support band-level uptime reporting today?
Honest answer: almost none do it natively. Most tools report uptime per monitor, not per severity band, so you compute band uptime yourself from incident exports.
Atlassian Statuspage gives you incident history and component-level statuses, which is enough raw data to derive bands, but no built-in band report. Better Stack and Hyperping store incident timelines you can export and post-process. UptimeRobot's history is thinner, so reconstructing partial-degradation windows is harder. Instatus handles status pages well but assumes you bring your own uptime math.
StatusPulse included: probes detect protocol-level failures, and incidents are tagged by severity, so band uptime is a filter over incident data rather than a rebuild. What it does not do today is generate the per-band SLA report automatically; you export incidents and run the calculation yourself. If you want a one-click band report, none of these tools ship it yet.
FAQ
Should scheduled maintenance count against my SLA?
No, if you announced it at least 48 hours ahead with a start time and expected duration. Report it as its own band, excluded from incident uptime with explicit timestamps. If an actual outage happens inside your maintenance window, the incident counts; the exclusion covers only the window you announced, not the downtime you caused during it.
How do I backfill band-based uptime for past months if I only have a blended number?
You can't, reliably, unless your incident history includes severity and component impact per incident. A log that only says "down at 3:12, up at 3:40" gives you no way to reconstruct partial-degradation windows or affected fractions. Start band reporting now, record severity and impact when each incident opens, and backfill only months where the raw data exists.
What uptime percentage should I promise in my first SLA as a small SaaS?
Promise less than you can deliver. A 99.9% monthly commitment allows about 43 minutes of downtime, which one bad deploy can burn, and your remedy clause keys off that blended figure. Publish band definitions alongside the number so a full outage is visible in the breakdown, and avoid 99.95% or 99.99% until you have a year of incident data.
Does a degraded-but-up incident count as downtime for SLA credits?
Under band reporting, partial degradation is scored separately, weighted by the affected fraction you published during the incident. Whether it triggers credits depends on your contract: most blended SLAs count only full unavailability, so a slow endpoint or delayed webhooks usually does not. State explicitly in your SLA which band, if any, triggers credits, and keep the incident record auditable.