Notification fatigue: when everything pings, nothing does

Notification fatigue is what happens when the cost of ignoring a notification drops to zero. Once every message, mention, comment and automated update arrives the same way, people stop reading any of them properly — and the one that actually mattered is lost in the same stream as a deployment bot. It presents as a discipline problem and is almost always a volume problem.

Where notification fatigue comes from

• Automated posts into human channels: builds, monitoring, form submissions. Individually reasonable, collectively the majority of the traffic.
• Channel-wide mentions used as a politeness device rather than because everybody needs to act.
• Large channels where everybody is a member because nobody wanted to exclude anyone.
• Cross-posting the same message to three places in case somebody misses it, which guarantees everybody sees it three times.
• Default settings nobody has revisited since the workspace was created.

Cut at the source, then at the receiver

Personal notification settings are where most advice stops, and it is the weakest lever: it moves the cost onto each individual and decays as soon as somebody is busy. Cutting at the source is what works — automated traffic into its own channels that people visit rather than get pushed, channel-wide mentions reserved for things with an actual deadline, and membership trimmed so a channel is the people who act rather than the people who might be curious. One change at the source removes the same noise for everybody at once.

Agree what warrants interrupting someone out of hours, in writing, as a team. Without an explicit rule the default is set by whoever is most anxious, and everybody else quietly matches it.

What should still interrupt

1. Anything with a deadline inside the next few hours that the person alone can act on.
2. Incidents, with an agreed definition rather than a feeling.
3. A direct question where the asker is blocked until answered.
4. Nothing else by default — including most mentions, which can wait until the person next looks.

Measuring whether it worked

Count the notifications a representative person receives in a normal week before and after. If the number has not roughly halved, the changes were cosmetic. The second measure is response time to the things that genuinely matter: fatigue shows up as a slower response to the important message, not as a slower response overall, because people are still reading — they are just reading everything at the same low attention.

Ettex Chat keeps automated traffic separable from conversation, so routing bots into their own channels is a configuration rather than a project. The decisions that come out of a channel belong in a log somewhere durable, which also removes the reason people re-post things to make sure they are seen.