Downtime Notification Policy: How to Write One That Works

Write a downtime notification policy your team will follow: detection rules, severity, roles, channels, update cadence, maintenance notices and a template.

A downtime notification policy is a short written agreement that answers four questions before an outage happens: what counts as downtime, who gets told, through which channel, and how often they hear back. Write it down once and nobody improvises at 3 a.m. This guide walks through each part, with a template you can copy into your own docs.

It is written for small teams running production services: engineers who carry the pager, founders who promise uptime to customers, and agencies who answer for client sites. The same policy covers websites, APIs and background jobs.

Uptime Beacon
uptimebeacon.eu
Fewer False Alarms for On-Call Teams
Uptime Beacon confirms repeated failures across regions before opening an incident, helping your team distinguish downtime from a network blip.
Explore Uptime Beacon

TL;DR

  • Split the policy in two: alerts wake the on-call engineer, notifications inform everyone else and never page anyone.
  • Define downtime by confirmed, user-facing failure (several failed checks in a row, verified from a second location), not by a single failed ping.
  • Name an incident lead and a communications lead in advance, and give customers one place to look: a status page that does not depend on your main app.
  • Commit to an update rhythm, and say in every message when the next one will arrive.
  • Cover planned maintenance, resolution and the post-incident write-up in the same document, and add a GDPR step if personal data could be involved.

Table of Contents

Alert or notification: the distinction your policy rests on

Most policies go wrong by mixing two different things. PagerDuty’s incident response documentation draws the line clearly: an alert requires a human to act, while a notification is informational and should not wake anyone up. Build your policy around that split.

Alert Notification
Audience On-call engineer Team, stakeholders, customers
Goal Get someone to act Keep people informed
Channel Phone, push, SMS, Slack ping Status page, email, chat channel
Hours Any time, within the escalation rules Business-friendly, except severe incidents
Example Checkout API failed from two regions Maintenance starts Tuesday at 22:00 UTC

Every rule below fits in one of these two columns. If you cannot say which one a message is, it probably should not be sent yet.

Step 1: Define what counts as downtime

The most expensive sentence in any policy is “we notify when a check fails.” A single failed request can be a network blip between the monitor and your server, and a policy that treats it as an incident teaches your team to ignore the pager.

Google’s SRE guidance says effective alerting is timely, covers key user-facing functionality, focuses on symptoms rather than causes, and is actionable. Turn that into concrete rules:

  1. Monitor what users do. Check the login endpoint, the checkout API, the public homepage, not just whether a server answers a ping.
  2. Require consecutive failures. Decide how many failed checks in a row open an incident. Three at a one-minute interval behaves very differently from three at five minutes, so write the number down along with the interval.
  3. Confirm from a second location. If one region sees a failure and another sees success, the problem may be the path, not your service. Confirmation before paging is the single biggest cut in false alarms.
  4. Treat silence as failure for scheduled work. Cron jobs and background workers do not serve requests, so use heartbeat checks: the job pings a URL when it finishes, and a missing ping opens the incident.
  5. Watch the slow burners. Certificate and domain expiry are downtime you can see weeks ahead. Give them a notification, not an alert, until the final days.

Uptime Beacon is built around this logic: an incident opens only after several failures in a row, and on paid plans it is confirmed from a second location before anyone is paged. Checks run from five regions (EU Central, EU North, US East, US West and Asia Pacific). Our longer piece on monitoring that pages only for real failures goes deeper on tuning thresholds, and the one on heartbeat checks for cron jobs covers background work.

Step 2: Set severity levels and escalation

Not every outage deserves the same response. Keep the scale small enough that someone half asleep can apply it. Three levels are usually plenty for a small team.

Severity What it means Who is alerted Who is notified
SEV1 Core service down or data at risk for most users On-call engineer immediately, then secondary after a set wait Status page, customer email, leadership
SEV2 Major feature degraded or one region failing On-call engineer Status page, internal chat
SEV3 Minor issue, workaround exists Ticket, no page Internal chat only

Then write the escalation path as a timed sequence. For example: primary on-call is alerted at confirmation; if nobody acknowledges within 10 minutes, the secondary is alerted; after 20 minutes, the engineering lead is called. The exact minutes are yours to choose. What matters is that they exist on paper, and that acknowledging an incident stops the clock.

Tie severity to your service level objectives where you have them. If a 99.9% target gives you roughly 43 minutes of budget a month, your SEV1 definition should reflect the incidents that would burn it. See our guide to setting service level objectives that work for how to pick the numbers.

Step 3: Assign roles before you need them

Small teams tend to have one person doing everything during an outage: debugging, answering customers and updating the boss. That is how updates stop. Atlassian’s getting-started guide recommends predefined communication roles, a central status page, templates for common failure scenarios and established escalation paths so there is less manual coordination under pressure.

AWS’s Operational Excellence guidance goes the same way: it says the policy should assign an incident manager, a communications manager and a support role. On a team of two, one person can hold two roles, but the roles should still be named:

  • Incident lead: decides severity, directs the fix, declares resolution.
  • Communications lead: writes and sends every external update, and owns the status page.
  • Support contact: relays customer questions to the communications lead instead of interrupting the engineer fixing the problem.

The U.S. government’s September 2026 joint guidance on outage communications makes a related point: it suggests a single intake path for information, coordinated by a designated communications lead, so engineers are shielded from outside interruptions during the incident. It also recommends consistent messaging across internal and external channels to avoid confusion and speculation. If you are a freelancer or agency, you are the incident lead and the communications lead. Say so in the policy, and add a backup contact for when you are unreachable.

Step 4: Choose channels and match them to audiences

Different people need different channels. AWS’s guidance lists status pages, email, SMS, in-app notifications, workplace chat and social media as options and stresses that the plan should be adaptable and use resilient channels. A simple mapping:

  • On-call engineer: a push or phone alert from your monitoring tool, plus Slack or email as a record. Uptime Beacon sends email, Slack and webhook alerts.
  • Internal team: a dedicated incident channel in your chat tool, opened for SEV1 and SEV2.
  • Customers: a public status page for every incident, plus direct email for severe outages. Atlassian’s guide says to notify stakeholders and customers through a status page and email customers in severe outages.
  • Partners and large accounts: a direct email or call from a named person, if you have contractual commitments.

One rule saves teams more than any other: do not host your status page on the infrastructure it reports on. If your app and your status page share a failure, customers see nothing at all. Uptime Beacon’s status pages do not depend on the dashboard, so they stay reachable during an incident. If you are weighing designs, our status page examples show what to copy and what to skip.

Also test your backup channel. The 2026 joint guidance recommends regularly tested backup communication methods. If Slack is down, where do you talk?

Step 5: Commit to an update rhythm

Customers forgive outages more easily than silence. PagerDuty’s guide advises that you alert users as soon as an incident occurs, update them regularly and tell them when it is resolved. Atlassian adds that you should set update intervals and say when the next update will occur.

For a concrete benchmark, the ePAY payments knowledge base publishes an outage plan in which verified interruptions are announced within 12 hours, updates follow every 30 minutes while unresolved, and a resolution notice is due within 24 hours of the first notice. That is one organization’s commitment, written for payment processing. A small SaaS team will usually want a faster first notice. A reasonable starting point:

Moment Commitment
Confirmation Status page entry as soon as the incident is confirmed and the impact is known
During the incident An update at the interval you promise, even if the update is “still investigating”
Resolution A resolved notice, with the time service was restored
Afterward A short root-cause summary for SEV1 and SEV2

Pick intervals you can keep. A promise of 30-minute updates that you break at minute 90 does more harm than a promise of hourly updates you keep.

What every update should contain

AWS recommends templates that include the service impairment, the expected resolution time and the impact. Keep one template per common failure so nobody drafts from scratch:

Status: Investigating / Identified / Monitoring / Resolved Impact: What users cannot do, and which regions or features are affected What we know: One or two plain sentences, no internal jargon Next update: A clock time

Do not guess at causes or promise fix times you cannot back. Say what you know, say what you are doing, and say when you will speak again.

Scheduled maintenance needs its own rules

Planned downtime should never surprise anyone, and it should never page your on-call engineer either. The ePAY plan requires maintenance notice 72 hours beforehand with the date, time, expected duration and affected systems, followed by a completion notice, and treats extended maintenance that interrupts service as a full incident. That is a good model for the shape of the notice, even if your lead time differs.

Your maintenance section should state:

  • How much notice you give (choose a lead time that fits your customers, and state it as a number of hours).
  • What each notice contains: start, end, expected impact, affected components.
  • A completion message when work is done.
  • What happens when the window overruns: the maintenance becomes an incident and follows the severity rules above.
  • That monitoring is silenced for the window, so alerts do not fire for downtime you planned. Uptime Beacon supports maintenance windows on Starter and above.

After the fix: resolution and post-incident communication

Resolution is not the last step. Atlassian’s incident communication guidance says post-incident messages should acknowledge and empathize with affected users, explain what went wrong, the fix and the prevention measures, and be documented in a postmortem. PagerDuty recommends writing public postmortems when appropriate.

For a small team, a good post-incident note is five short paragraphs: what happened, who was affected, how long, why it happened, and what changes now. Write it within a few days while details are fresh, and give it an owner and a deadline in the policy. Our incident response playbook includes a first-hour checklist you can pair with this.

When downtime involves personal data

Most outages are availability problems, not data breaches. But if an incident exposes or loses personal data, GDPR adds a clock. The European Data Protection Board says the controller must notify the supervisory authority without undue delay and, where feasible, within 72 hours of becoming aware, unless the breach is unlikely to risk individuals’ rights and freedoms. For breaches that affect people in several Member States, the same guidelines point to the supervisory authority of the controller’s main or single establishment as the lead.

Add one line to your policy: the incident lead decides, at the first sign of possible data exposure, whether the breach process starts, and names the person who files. If you serve EU users, note that GDPR applies to EU companies wherever they process data, and to non-EU companies offering services to people in the EU. This is a place to involve your legal adviser or data protection officer, not to improvise.

A policy template you can copy

Paste this into your runbook and fill in the brackets:

  1. Scope: Services covered: [website, API, cron jobs, certificates and domains].
  2. Detection: An incident opens after [N] consecutive failed checks at [interval], confirmed from a second location.
  3. Severity: SEV1, SEV2 and SEV3 as defined in [link].
  4. Alerts: Primary on-call is alerted at confirmation; secondary after [X] minutes without acknowledgment; lead after [Y] minutes.
  5. Roles: Incident lead [name]; communications lead [name]; backup [name].
  6. Channels: Status page [URL]; customer email list [name]; internal channel [name]; backup channel [name].
  7. Cadence: First public notice within [X] minutes of confirmation; updates every [Y] minutes; each states the next update time.
  8. Maintenance: Notice [N] hours ahead; completion notice afterward; monitoring silenced during the window.
  9. Resolution: Resolved notice at restoration; post-incident summary within [N] days for SEV1 and SEV2.
  10. Data: If personal data may be affected, the incident lead consults [DPO or legal contact] immediately.
  11. Review: Policy tested and revised every [quarter], and after any SEV1.

Put the policy to work with Uptime Beacon

A policy only helps if your tooling follows it. Uptime Beacon’s Free plan, with no card required, gives you 5 monitors checked every 5 minutes from EU Central, one alert channel (email or Slack) and one status page, enough to run the detection and customer-notification parts of this policy on a small project. Paid plans add second-location confirmation before an incident opens, and maintenance windows from Starter up. Starter is €14.99 per month, Pro is €39.99 and Business is €89.99; extra monitors and expiry watches are available as add-ons from €5 per month. See the plans or look at our API and cron monitoring for SaaS teams.

Our perspective: the policy starts with the page you do not send

We have spent more than 30 combined years running production in banking, fintech, telecom and SaaS, and the pattern repeats: teams do not fail their policy because it is missing a section. They fail because the pager fires too often to be trusted. When engineers learn that half the alerts are network blips, they acknowledge slowly, and the real incident gets the same slow response.

That is why we put the alert threshold first. Fewer false alarms means that when a page does arrive, it is believed, and the notification steps that follow (status page, customer update, next-update time) start earlier. Everything else in a policy depends on that trust.

Our perspective: the policy starts with the page you do not send — overview diagram

The second thing we have seen is that the policy is read mostly by the tired. Keep it to a page. Use a template for every message. Name people, not teams.

The third is newer. Many of our users now ask an AI assistant to look at a failing monitor before they open a laptop. Over MCP, on every plan including Free, an assistant such as Claude, ChatGPT or Cursor can read monitors, incidents and check history and diagnose a failure region by region. With Allow changes turned on for a team, it can also acknowledge incidents, pause monitors and start maintenance windows. Nothing can be deleted over MCP. We suggest your policy says explicitly whether the assistant may act or only read. Our posts on AI incident response and AI for monitoring explain where we draw that line.

— Aaditya Parashar

FAQ

What is the difference between an alert and a notification?

An alert needs a human to act and can justify waking someone up. A notification is informational, such as a maintenance notice or a status page update, and should not page anyone. Keeping the two separate is what keeps on-call engineers responsive.

Should I notify customers during scheduled downtime?

Yes, in advance. Send a notice that gives the start, expected duration and affected systems, then a completion message afterward. Silence your monitoring for the window so planned work does not trigger alerts, and treat an overrun as a normal incident.

How often should I update customers during an outage?

Pick an interval you can reliably keep and state it in the policy. Many teams promise an update every 30 to 60 minutes while an incident is open. Always end each message with the time of the next update, even if the news is that nothing has changed.

Does a downtime notification policy need to mention GDPR?

Only where an incident could involve personal data. Most outages are availability events, but if data may be lost or exposed, the controller has to assess whether to notify the supervisory authority, generally within 72 hours of becoming aware. Name who makes that call and who to consult.

Who should own the policy on a very small team?

One person, usually whoever carries the pager. A solo developer or freelancer holds all the roles, so the policy mainly records the thresholds, the channels and the message templates, plus a backup contact in case you are unreachable.

Sources

Put your site on it. Sleep through the blips.

Add your first monitor now. If your site is fine, you will not hear from us.

Monitor it free
Sign up with GoogleOpening Google…
Free plan, no cardSign up with Google or email