Inbox triage for LinkedIn: whose turn is it?

Open LinkedIn on a Monday with sixty active conversations and the first decision — which thread to read — is made for you by the worst possible rule: whoever wrote most recently is at the top. Recency is not urgency. The message that arrived twenty minutes ago is often “thanks!”, and the one that needs you today slid off the first screen four days ago.

Unread does not help either. You read a message on your phone in a taxi, did nothing about it, and it is now indistinguishable from a message you handled.

The question worth sorting on

There is only one question you are really asking as you scan that list: is anything here waiting on me?

It sounds like the same thing as “is this conversation finished”, and it is not. That distinction is worth dwelling on, because getting it wrong produces a feature that looks right and is useless.

Ask whether an exchange is finished and almost every prospecting thread stays open — something in it is always unresolved, a deal never formally died, nobody said no. Sort on that and your “needs attention” view contains fifty-five of your sixty conversations, which is the same as containing none of them.

Ask whether anything is waiting on you and the list collapses to the handful where the next move is genuinely yours:

• they asked a question you have not answered;
• they asked for something — a deck, a price, an intro — and you have not sent it;
• they proposed a time and you have not confirmed it.

Everything else — including the live, healthy, going-somewhere deals — steps aside. A thread whose next move belongs to them is done for now, however alive it is. That is the part that feels wrong until you try it: “they said they’d come back Monday” is a finished piece of work, not an open one.

When we changed the question from the first form to the second, on the same set of threads, the readings moved exactly where they should: we answered and are waiting went from 0.25 to 0.82, they said they would come back Monday from 0.06 to 0.84, they are checking with a partner from 0.08 to 0.84 — while the ones where the ball genuinely is in our court barely moved at all (their unanswered question, 0.04; a deck they asked for and never got, 0.07). The view emptied of everything that was not ours to do.

The bit that makes it safe to hide things

A view that hides conversations is only acceptable if hiding is reversible without you remembering to check. Two rules do that.

A new message brings a thread straight back. Obvious, and not enough on its own — plenty of threads go quiet precisely because the other person is waiting for a date to arrive.

A thread that steps aside names when it should come back. “Ask me in the spring” is information. So is “I’ll check with my partner this week” and “let’s talk after the conference”. A thread that steps aside on one of those should reappear when the moment arrives, without you writing it in a notebook.

One detail decides whether that is useful or infuriating: the return has to land past the word given, not on it. Someone who said they would come back Monday must not resurface Monday at nine in the morning, before they have had the day. They come back Monday evening, or Tuesday. Resurfacing on the day turns a smart feature into a nagging one.

Your own reply is a fact, not a reading

When you answer a thread, it leaves the pile. That is not something to ask a model about — it is simply true, and it is recorded as a fact. Same in reverse: writing back into a settled thread revives it.

This matters more than it sounds. A system that re-reads a conversation every time you send a message is paying for a reading to work out something it already knows, and it will occasionally get it wrong and tell you that you left a question open two messages after you answered it. Facts should be recorded. Only genuinely ambiguous things should be read.

And a thread nobody has looked at yet is not a verdict. An unread conversation should be absent from a “waiting on you” view rather than assumed to be handled — silence from a classifier is not the same as a clean bill of health.

The other half: before you press send

Triage gets you to the right thread. The second failure mode is the message you then write, and it is the one nobody catches because you are the only person who reads it before a prospect does.

Four things genuinely go wrong in a LinkedIn reply, and all four are checkable before you send:

An unfilled variable. {firstName} on its way to a hundred people. This is a regex, not an opinion, and it should outrank everything else.

The wrong person or the wrong language. You had two threads open and answered in the other one’s language. Below about five words this cannot be judged — the language of “Ok thanks” is a coin flip — but above it, it is reliable.

A contradiction with the thread. You just quoted a different price than you did three messages ago.

A question of theirs left open. With one hard rule attached: this can only fire when they spoke last. Once you have written last, the thread is not waiting on you, whatever they asked earlier — you had your turn. We found this live: a contact asked “is it a setter?”, we answered, we followed up with “give it a try”, and the draft after that was told it had left their question open. The model was reading their last message where the thread’s state was what mattered.

What should not be checked is anything to do with style. Length, tone, politeness, thoroughness: none of them is a defect. A curt “Yes, Thursday 2pm” is a perfect reply, and a tool that suggests padding it out will be switched off by the end of the week. The moment you ask a model “how good is this message”, all four come back in through the side door.

What a triaged Monday looks like

Six conversations where something is genuinely yours to do, each with a line saying what. Fifty-four out of the way, each carrying its own return date. One badge beside the send button confirming the draft was read against the thread and nothing looks wrong.

The inbox did not get smaller. The decision about what to read first stopped being yours to make.