Most channel choices take about two seconds and get made by whichever window is already open. That works for the majority of messages, because the majority of messages are small. It fails on the ones that matter, and it fails in a predictable direction: the decision that needed a written record gets made on a call, and the question that needed thirty seconds of back-and-forth arrives as nine paragraphs.
One variable settles most of it. Do you already know what you are asking?
A call wins when you do not have the question yet
If you cannot state the question in a sentence, you do not have a question. You have a mess, and what you need is the other person's questions to find yours. That is the case a call is unbeatable for. Ten minutes of someone asking "wait, why does that have to be true?" locates the real problem faster than any amount of writing, because you cannot write your way past a blind spot you cannot see.
The other clean case is when the emotional content matters more than the record. Bad news. A disagreement that has gone tense. Anything where tone carries meaning the words do not, and where the other person's ability to interrupt is part of the point. Delivering that in writing is not efficiency, it is avoidance with better formatting.
The failure is the record. A call's output is what each person remembers, and two people remember different decisions with equal confidence and no way to adjudicate. The fix is a written summary afterwards, and the reason it rarely gets written is that nobody wants to do the work twice. A call also charges both people the same hour regardless of how much of it either needed, and it requires both awake at once — which across enough time zones means somebody takes it at an hour that degrades the judgment you called for.
Email wins when the context is already shared
Email is very good at short, bounded, low-context exchanges, and that is most correspondence. The recipient already holds the background, the question fits in a sentence, the answer fits in a sentence. Confirming a date. Checking whether a thing shipped. Passing a link with one line about why. Nothing beats it there, and reaching for something heavier is a mistake in the other direction.
What email cannot do is carry context, because it has no structure to carry it in. The ask ends up somewhere in the fourth paragraph and the reader has to work out which sentence was the question. There is no marking for what you know versus what you assumed, so a reader cannot weigh the parts differently even when they should. And length is the only signal a long email sends before it is read, which is why the messages needing most from a reader are the ones most likely to sit until the weekend.
A newer failure is worth naming. A fluent draft written by an assistant reads confident and says less than it appears to, and the reader cannot tell which sentences the sender actually stands behind — the argument for thinking hard about when not to hand a message to an AI at all.
A briefing wins when there is context to move and one decision to get
The briefing's case is narrow. There is genuine context the other person does not have, you want one particular thing decided, and the two of you are not going to be awake at the same time anyway. A briefing is a structured package rather than prose — an explicit ask with a response shape (opinion, decision, review, info or fyi) and a stated urgency, a bounded context brief, assumptions marked as stated or inferred, options with their status, and the questions still open.
The structure does one job: it removes the round trip. What the recipient would otherwise have had to ask for is already there, so they can answer on first read instead of sending three clarifying questions and waiting a day for each.
The trade is that a briefing is expensive to write and cheap to read — the right direction when you are the one asking for the favour, and the reason to be honest about where it is the wrong tool.
- A two-line question. Filling structured fields around one sentence is absurd. Send the email.
- Anything urgent. A briefing is asynchronous and nothing makes it faster. There is no push to the recipient's assistant — email is the notification channel, and their assistant only sees the package when they ask it to check. If you need an answer within the hour, that is a phone call, or it is not as urgent as it feels.
- A conversation that needs to be a conversation. Four turns of clarification is four waits on someone's attention. Some problems are cheaper to talk through, and choosing a written channel to avoid that is the same avoidance in a different costume.
- Thin status. You can see whether the latest package has been seen — their assistant pulled it, or they opened the link — and nothing more. No timestamp, no count. The notification email is plain text with no tracking pixel, so an unopened link and an unread message look identical from your side.
- One recipient. Threads are two-party. If four people need to weigh in, that is four threads and four replies to reconcile yourself.
Choosing in ten seconds
- Can you state the question in one sentence? If not, call.
- Does the answer depend on how the other person feels about it? Call.
- Does the recipient already hold the context, and does the whole exchange fit in two sentences each way? Email.
- Is there real context to move and one specific decision to get, across schedules that do not overlap? Briefing.
- Do you need it within the hour? Call, or admit that you do not.
- Are you picking a written channel to avoid a conversation you are dreading? That is the tell. Take the call.
The last one is worth sitting with, because meetings and asynchronous decisions fail in opposite directions, and most people are reliably consistent about which failure they prefer rather than which one the message in front of them calls for.
Two channels is often the right answer
The best correspondence often uses more than one, in a deliberate order.
Brief first, then call, and the call drops from ninety minutes to fifteen because nobody spends the first hour establishing what the question is. Or call first, then write, because the call finds the question and the writing makes the answer durable — the moment to produce a decision record, while the reasoning is fresh and the losing options still look like options.
What makes either sequence work is naming the job before picking the tool. "I need their questions" and "I need their answer" are different requests, and almost every badly chosen channel is one of the two mistaken for the other.
If the job in front of you is the second one — real context, one decision, no shared calendar — connect your assistant and let the structure do the work the prose was failing at.