Back to blog
6 min read
AlertingGuide

Which alert channel actually wakes you at 3am

An honest guide to on-call alert reliability: the 3am test, the trade-offs of email vs chat vs push vs SMS vs voice, and how to get a real page today.

NK

Nabin Khair

Founder

There is a test every on-call engineer eventually runs, usually by accident, usually at the worst possible hour. Your phone is on the nightstand, face down, Do-Not-Disturb on, and something is failing in production. The only question that matters is whether the alert you set up actually gets through the silence and wakes you. Everything else — the dashboard, the runbook, the post-mortem — happens after you're awake. The page that wakes you is the only link in the chain that runs while you're unconscious, and it's the one people tune the least.

I build Tallwatch, so I'll say up front where we land before I give you the honest ranking: we ship seven alert channels and they're great for routing and chatops, but we do not have native SMS or phone-call paging yet. That shapes my advice rather than excusing it, and I'll tell you exactly how to get a reliable 3am page with what ships today.

First, the ranking. There's a view that gets repeated in every 2026 on-call discussion I've seen, and after years of carrying a pager I think it's basically correct: a phone call is the only actually reliable notification type. Every other one can be delayed. Let's go through them in order of how much I'd trust them to wake someone.

Email: awareness, never a page

Email is a record, not an alarm. It's perfect for the audit trail — what fired, when, what the consensus looked like — and useless as the thing that wakes you. It batches, it gets filtered, it sits behind notification settings that vary by client and by phone. Nobody's mail app reliably overrides Do-Not-Disturb at 3am, and you shouldn't ask it to. Use email so the morning has a paper trail. Never use it as the page that has to wake someone.

Chat — Slack, Discord, Teams: where the team gathers, not where the page lands

This is the category most teams accidentally over-trust, so I want to be careful. Chat is genuinely excellent — for the waking hours. During business hours, a Slack or Teams alert lands in the channel where the responders already live, threads the discussion next to the incident, and pulls the right people together with zero context-switching. That's real value and it's most of what chatops is for.

But chat depends on a push notification arriving on a phone whose owner has not muted that app for the night — which, if they're trying to sleep, they have. The notification can also be delayed by the chat provider's own delivery path; these aren't designed as paging systems. So my rule is simple: chat is for awareness, not for the page that must wake someone. Route incidents to Slack, Discord, and Teams so the team can coordinate. Do not make a muted chat app the last line between an outage and a sleeping engineer.

Push notifications: better than chat, still beatable by sleep

A dedicated push from a monitoring or paging app is a step up from a chat notification — it can be configured as time-sensitive and is more likely to punch through focus modes. But "more likely" is doing a lot of work. Push still rides the same OS notification system, and a tired person who has aggressively configured their phone for sleep can still miss it. Useful, not guaranteed.

SMS: feels reliable, often isn't

SMS has a reputation for reliability it doesn't fully earn anymore. It frequently arrives, and occasionally arrives late — carrier queuing and delivery delays are real and you don't control them. It also doesn't inherently override silent mode on every phone. It's a reasonable second hop in an escalation, but I wouldn't make it the sole thing standing between you and a sleeping responder.

Voice: the one that actually wins

A phone call is the only channel that, by default, most phones treat as important enough to ring through Do-Not-Disturb (especially on a repeat call), and a ringing phone demands a decision in a way a buzz never does. It's the channel that passes the 3am test. If a page absolutely must wake a human, it should end in a phone ringing.

So how do you get that today, with Tallwatch?

Here's the candid part. Tallwatch ships seven channels — email, Slack, Discord, Telegram, Microsoft Teams, signed webhooks, and PagerDuty — and they cover awareness and chatops well. They do not include native voice or SMS. So if I stopped here and told you to wire up Slack and call it on-call, I'd be selling you the exact mistake this post is about. Instead, here are the two ways to get the real page today:

Route the critical escalation level through PagerDuty. In Tallwatch you build escalation policies with ordered levels, each with its own timeout, and acknowledging stops the cascade. Point a top level at the PagerDuty channel, and PagerDuty delivers the phone call. You get Tallwatch's consensus, on-call rotations, and escalation; PagerDuty supplies the voice page we don't.

Or use a signed webhook into a voice/SMS service you run. Tallwatch sends signed webhooks (with programmable payload templates on paid plans), so the critical level can fire a webhook into a voice or SMS provider you control, and that provider places the call. More setup, but it's yours end to end.

Either way, the pattern is the same: chat channels for awareness, a voice-capable level for the page that must wake someone. Build your escalation so the early levels nudge the channels people watch while awake, and the late, must-not-be-missed level reaches a phone that rings.

The reliability you can't buy with a channel

One more thing, because it's the part people skip. The most reliable channel on earth is worthless if it cries wolf. If half your pages are false, your responder will mute the channel, ignore the ring, or build a rule to silence it — and now your perfect voice page wakes no one because nobody believes it anymore. Alert reliability is the channel and the trust in it, and the second half is where most tools quietly fail.

This is why Tallwatch pages by consensus: an incident only opens when at least two checked regions agree the target is down. One region having a bad network hour doesn't reach your escalation policy at all. So the page that finally rings the phone is one your responder has reason to trust — which is the only thing that keeps a 3am call from becoming a 3am habit of ignoring calls.

Get both right: a voice-capable escalation level for the page itself, and consensus so the page is worth answering. That's the on-call setup that actually wakes the right person for the right reason. If you want to see how the flat-priced, bundled version of this works — and why I think paying per seat for the coordination tool is backwards — start there.

Start free.

Keep reading