The most common visibility setup we see in the wild is a Slack channel called #shipping-alerts with four hundred unread messages and a mute icon next to its name. Everyone agrees the alerts are important. Nobody has opened the channel since March.
That’s not a discipline problem. It’s a design problem, and it has a design solution.
An alert nobody acts on is worse than no alert: it teaches the team to look away.
Severity that means something
Most alert pipelines have severity levels; almost none have severity definitions. Ours are blunt on purpose:
- High — a promise to a customer is now at risk. Interrupt someone.
- Medium — the plan changed, but no promise is broken yet. Same-day attention.
- Low — useful context. Batch it into the daily digest and never interrupt anyone with it.
The test for every rule is simple: if this fired at 2 p.m. on a Tuesday, what should a specific human do? If the honest answer is “nothing, really,” it’s a Low, no matter how dramatic it sounds.
Route to an owner, not a channel
Channels diffuse responsibility; owners concentrate it. Every shipment in Windrose has an owner, so every alert lands on a person — with the shipment, the reason, and the delta in the first line. The channel still exists for visibility, but the notification goes to the one human who can act.
This one change does more for acknowledgement rates than any amount of “please watch the channel” in a team meeting. People respond to their name. They do not respond to @here.
Respect the watch schedule
Freight moves at 3 a.m.; your team shouldn’t have to. Quiet hours hold everything below High until morning, and the overnight Lows arrive as one tidy digest instead of eleven pings. If something genuinely can’t wait, it was a High all along — and Highs are rare, which is exactly what keeps them loud.
Measure acknowledgement, not sends
The health metric for an alert pipeline is not how many alerts it sends — it’s how many get acknowledged, and how fast. We watch two numbers: acknowledgement rate per rule, and median time-to-ack. Across our customers the median is twelve minutes, and any rule that falls under a 60% ack rate gets reviewed, demoted, or deleted.
Deleting an alert rule feels like negligence and is usually hygiene. A pipeline of few, loud, owned alerts beats a firehose every single week of the year. That philosophy is built into how exceptions work in Windrose — but it’ll improve whatever system you run today.