Reply all: how one message becomes four hundred

The options around reply-all storms are set out side by side below, with the conditions that genuinely favour one over the other.

The mechanism needs only three ingredients

A storm requires a large distribution list, permission for members to reply to it, and one message sent to it by mistake. The first reply asking why they received this goes to everyone, which demonstrates to everyone that replying to everyone is possible.

Subsequent messages divide into requests to be removed, requests to stop replying, and observations about how many messages have now arrived. Each category generates the next, and the volume peaks within the first hour before decaying over the rest of the day. The shape is consistent enough across organisations that mail administrators recognise it immediately and dread it accordingly.

Asking people to stop is the accelerant

A message telling everyone to stop replying to all is itself a reply to all, which is the joke and also the mechanism. It also raises the salience of the thread for people who had ignored it, which pulls in participants who were not previously involved.

On hold for the fourth time, the correct response is silence, and silence is difficult because each individual believes their reminder will be the one that works. This is a coordination problem rather than a stupidity problem, since every individual action is locally reasonable and collectively harmful. It resolves the same way most coordination problems do, which is by somebody with administrative access changing the rules.

The technical fix is old and simple

Mail systems allow a distribution list to be configured so that only named senders can post to it, which prevents the entire category. Moderation, reply-to redirection and message rate limits are all standard features and all predate the problem by decades.

The reason they are not applied in advance is that large lists are usually created for one announcement and never reviewed afterwards. Most organisations therefore apply the fix immediately after their first storm, which is a very expensive way to schedule a configuration change. Auditing which lists accept replies from anyone is a half-hour task that prevents a full-day event.

Why the cost is larger than it looks

Several hundred messages to several thousand mailboxes is a genuine load on mail infrastructure, and storms have taken systems down. The larger cost is attention, since every recipient interrupts whatever they were doing at least once, and many do so repeatedly. Filing, muting and deleting the thread is itself work, and it is work performed simultaneously by everyone in the organisation.

Three clicks later, the event also trains people to ignore messages sent to large lists, which degrades the channel used for genuine announcements.

That last effect persists long after the thread dies and is the part nobody measures.

The near relatives of the storm

Out-of-office replies configured to respond to list mail can produce loops, which is why mail systems suppress automatic replies to lists by default. Read receipts requested on a message to thousands of people generate thousands of receipts, which is the same event with a different trigger.

On hold for the fourth time, recall requests are perhaps the purest form, since a recall notice is a second message announcing the existence of the first. Group chat has inherited the pattern in the form of large channels where everyone reacts to a mistaken post. The underlying condition is always the same, which is a broadcast channel that anybody can broadcast on.

Plenty of this is legal, which is rather the point — the complaint is that it is allowed, not that it is criminal.

What to do while it is happening

Mute the thread rather than deleting messages, since deletion is a task that repeats and muting is a task that does not. Do not reply, do not reply asking others not to reply, and particularly do not reply to point out the irony.

If you have administrative access, disable replies on the list first and communicate afterwards, because the order matters enormously. A single message from a senior person, sent individually to the loudest participants rather than to the list, ends it faster than anything public. Afterwards, audit the lists, because the storm has just told you exactly which one was configured wrongly.

Side by side

The takeaway

The only correct contribution to a reply-all storm is nothing at all.

It is not you being fussy. It is genuinely badly made.