The one-sentence rule

Page if a human has to do something now to stop customer harm, data loss, or an active security issue.
Notify if someone should know soon, but sleep is more important than speed.
Ticket if it can wait for the next working day without making the problem worse.

If you cannot say which bucket an alert belongs in, it is not ready to ship.

Examples that usually do not

CPU is high but the product still works.

A single region looks slow and others are fine — investigate, do not necessarily wake the company.

Staging is red.

A disk warning at 70% with days of headroom.

"Anomaly detected" with no action attached.

Those can go to Slack, email, or a daytime queue. They should not climb an escalation policy.

Channels are not severities

Slack is a great place for awareness. It is a bad exclusive path for "wake a sleeping human." Push, phone bridges (via a signed webhook into Rootly or similar if you need voice), and devices that break through Focus modes exist because chat apps are designed to be polite.

Tallwatch ships email, Slack, Discord, Teams, Telegram, and signed webhooks — and is honest that native SMS and voice are not in the box yet. Whatever tools you use, match the urgency to the channel, not the other way around.

A filter that keeps you honest

Before you enable a page, write the first three steps the on-call person will take. If you cannot, you do not have an actionable alert — you have a notification wearing a pager costume.

Then look at last month's pages. Anything that was acked and closed with "no action" should be demoted. Your future selves will sleep better.

Agent

Page as markdown

Raw agent-readable copy for this route. Copy it into a model, or browse the site-wide agent index at /llms.txt.

---
title: "When should you page someone vs just notify the channel?"
description: "Not every red alert should wake a human. Here is a simple rule for what deserves a page, what belongs in Slack, and what should wait until morning."
canonical: https://tallwatch.com/blog/when-to-page-vs-notify
url: https://tallwatch.com/blog/when-to-page-vs-notify
llms_url: https://tallwatch.com/llms.txt
updated: 2026-07-15
---
# When should you page someone vs just notify the channel?

Not every red alert should wake a human. Here is a simple rule for what deserves a page, what belongs in Slack, and what should wait until morning.

Published 2026-07-15 · by Nabin Khair
Tags: Alerting, On-call

On-call dies in two ways: nobody gets paged when they should, or everybody gets paged when they should not. The second is more common, and it is almost always a sorting problem.

## The one-sentence rule

**Page** if a human has to do something *now* to stop customer harm, data loss, or an active security issue.
**Notify** if someone should know soon, but sleep is more important than speed.
**Ticket** if it can wait for the next working day without making the problem worse.

If you cannot say which bucket an alert belongs in, it is not ready to ship.

## Examples that usually deserve a page

- Login or checkout is down for everyone (confirmed, not a single-probe blip).
- Data is being corrupted or not writing.
- A security boundary just failed open.
- A dependency you cannot route around is dead and customers are blocked.

## Examples that usually do not

- CPU is high but the product still works.
- A single region looks slow and others are fine — investigate, do not necessarily wake the company.
- Staging is red.
- A disk warning at 70% with days of headroom.
- "Anomaly detected" with no action attached.

Those can go to Slack, email, or a daytime queue. They should not climb an [escalation policy](/blog/what-is-an-escalation-policy).

## Channels are not severities

Tallwatch ships email, Slack, Discord, Teams, Telegram, and signed webhooks — and is honest that [native SMS and voice are not in the box yet](/blog/which-alert-channel-wakes-you-at-3am). Whatever tools you use, match the *urgency* to the *channel*, not the other way around.

## A filter that keeps you honest

Before you enable a page, write the first three steps the on-call person will take. If you cannot, you do not have an actionable alert — you have a notification wearing a pager costume.

Then look at last month's pages. Anything that was acked and closed with "no action" should be demoted. Your future selves will sleep better.

When should you page someone vs just notify the channel?

The one-sentence rule

Examples that usually deserve a page

Examples that usually do not

Channels are not severities

A filter that keeps you honest

Keep reading

What changed after we stopped paging on warnings

How to cut false downtime alerts without waiting forever to get paged

Webhook alerts that silently fail: how to know your pager is broken

Page as markdown