A thread’s gone quiet, and you’re wondering whether it’s too soon to chase. It’s usually later than that.
Two business days for a client email that asked a question. Five for one that asked somebody to go and find a document. Same day, if a date you can’t move sits inside either window.
Those are defaults rather than rules. They’re shorter than most people’s instinct because replies are extremely front-loaded. The reply that was going to come has usually already come. Wait a polite week and you spend six of those days in the part of the curve where almost nothing happens, then chase at the moment the thread has gone properly cold.
- The typical reply is much faster than the average reply. A small number of very late answers drag the average up, and your instinct is calibrated on the average.
- The right interval comes from how fast that particular person normally gets back to you, rather than from a general rule of etiquette.
- Weekends, evenings and out-of-office days sit outside the wait. Count them in and a two-day interval quietly becomes five.
- The point where silence starts to mean something is roughly ten times later than the point where a follow-up starts being useful. Confusing the two is the common error, in both directions.
This piece is about the interval alone. What the follow-up message should say is the follow-up email that gets answered. Whether a run of them reads as nagging is a question about the pattern rather than the gap, and it’s settled in following up without nagging. Money keeps its own clock, set by your terms rather than by anybody’s attention, and that one is following up on an unpaid invoice.
The curve almost nobody has actually seen
Reply times pile up hard at the start. Then they trail off for a very long way. The shape is stable enough that it’s been measured the same way for twenty years.
The clearest description of it is in Pauses and Response Latencies, published in the Journal of Computer-Mediated Communication in 2006 by Yoram Kalman, Gilad Ravid, Daphne Raban and Sheizaf Rafaeli. They took three unrelated archives, totaling more than 150,000 responses:
- a university forum
- a commercial question-and-answer service
- 15,815 email responses written between 1998 and 2002 by 144 employees of the Enron Corporation, whose mail became public through a Federal Energy Regulatory Commission investigation
Three archives, three technologies, three populations, one shape. The latencies followed a power-law distribution, “such that most of the responses (at least 70%) were created within the average response latency of the responders, while very few (at most 4%) of the responses were created after a period longer than 10 times the average response latency”. The authors say this held “across diverse user populations, contexts, technologies, and average response latencies”.
The first half of that sentence is the surprising part. At least seven replies in ten arrived before the average. An average is supposed to sit in the middle of things, and here it sits well past most of the replies, because the handful that took a month pull it upward. So the number in your head for how long a reply usually takes is already too long. Everything you calculate from it inherits that error.
Kalman and Rafaeli revisited the same dataset in 2011. For the corporate mail, they put it in plain numbers: “more than 85% of the responses were sent within about one day, and that only 3% of the responses were sent after 12 days or more”.
That’s an archive from the turn of the century, so the fair question is whether it still describes anything. It does. The largest modern measurement is Evolution of Conversations in the Age of Email Overload, a 2015 paper by Farshad Kooti, Luca Maria Aiello, Mihajlo Grbovic, Kristina Lerman and Amin Mantrach, built on 16 billion Yahoo messages. It found the same front-loading: “More than 90% happen within a day of receiving the message, and the most likely reply time is just two minutes. Also, half of the replies are within 47 minutes of receiving the message.”
Both of those describe replies that came. How many messages went unanswered altogether sits outside what they measured, and that’s the right caveat to carry. What they establish is narrower and more useful. A reply coming through the normal route comes early. A thread that’s quiet on day three has already left the region where quiet resolves itself.
The line to chase on and the line to stop on
The 2006 paper also does something unusual for academic work. It proposes a threshold you can actually apply. It defines online silence as the absence of a response “after a period of ten times the average response latency”, a line that “yields a better than 95% confidence level that a response is not likely to occur in the future”.
So there are two lines on the same curve, ten times apart.
The normal window is roughly how long that person usually takes to answer you. Inside it, everything is running the way it should, and a follow-up is just noise landing on top of a message they were going to get to anyway.
The silence line is ten times that. Past it, the research says you’re more than 95 percent likely to be waiting on something that will never arrive. For the Enron mail, where a normal answer took about a day, that line fell at twelve days.
Almost every bad follow-up interval comes from picking the wrong line. Wait until the silence line and you’ve followed up too late to be useful, because you spent the whole productive middle of the curve waiting. Fire inside the normal window and you’ve interrupted somebody who was already handling it.
The interval you want sits in between, much closer to the first line than the second. Once the normal window has clearly closed, the reply is in the tail, and the tail is exactly where a second message earns its keep. It works by arriving back at the top of a pile on a better day.
That gives a working rule that adapts to any relationship. Find how long this person normally takes with you, and follow up at about twice that. Treat ten times that as the point where the channel itself has failed and something other than another email is the move, which is chasing a client who will not reply.
Count business days and start when they read it
Two days on a calendar and two days in somebody’s inbox are different lengths. The same study that measured enterprise reply behavior measured what the calendar does to it, and the effect is much larger than people allow for.
In Characterizing and Predicting Enterprise Email Reply Behavior, presented at the 2017 ACM SIGIR conference, Liu Yang, Susan Dumais, Paul Bennett and Ahmed Hassan Awadallah worked through the Avocado collection, 938,035 emails from 279 accounts of a defunct technology company. On timing they found that “the reply rate for emails sent on weekdays is around 7%, but that rate drops to 4% on weekends”. Weekday mail was answered in under an hour. Weekend mail ran far behind it: “emails sent on the weekend have 13 (for Sunday) or 30 (for Saturday) times longer reply latency”.
Thirty times is a different order of thing. A message sent on a Saturday morning starts existing, as far as its recipient is concerned, on Monday.
Time of day pushes the same way on a smaller scale. In that archive, mail sent in the morning or afternoon had a median reply time of under an hour. Mail sent in the evening or overnight took more than seven.
Three practical corrections follow, and each one is free.
Count business days only. A Thursday afternoon request followed up on Monday morning is a two-day interval. The shorter number is the one that should govern how you write.
Start the clock from when it plausibly reached them. If it went out at 5:40 on a Friday, Monday morning is hour zero.
Add their days off to yours. An out-of-office reply extends the interval by the days it names. It doesn’t restart it, and the request is still sitting there waiting to be read. The day somebody comes back from vacation is also the worst possible day to land in their inbox, so add one more.
All of this assumes the delay is personal rather than structural. In the run-up to a filing deadline everybody is slower, your clients included, and the six quiet months after an extension run on a completely different rhythm from the six weeks before one.
What you asked for sets the interval
The single biggest thing separating a two-day wait from a two-week one is what you asked them to do. The same SIGIR study puts numbers on it from two directions.
Messages that contained a request were “almost twice as likely to receive replies as those that do not, 14.81% vs. 7.45% respectively”. The median wait on them ran to 81 minutes rather than 55. The authors’ explanation is the obvious one and worth taking seriously. A request is attached to a task. Before anybody can reply they have to find a file, look something up, or ask somebody else, and that takes longer than the ninety seconds they had.
The other direction is who you are. In that archive, “people are 3.4 times more likely to reply to internal emails (7.76%) than external emails (2.26%)”, and internal mail came back roughly twice as fast, at a median of 66 minutes against 135. Every message you send a client crosses that boundary.
Those are one company’s mail. The absolute rates count everything, including the announcements nobody would ever answer, so the levels are theirs rather than yours. Keep the ratios, because they stack. Client mail that asks for something is external and a request at once, so the normal window for it runs several times wider than your general sense of how fast email moves. Time a document chase off your intuition and you’re timing it off the wrong distribution.
Translated into intervals a small practice can use:
A question they can answer from memory. A yes or no, a preferred date, a spelling, a confirmation that something is right. Normal window is same day to next morning. Follow up after two business days.
A request for something they have to go and find. A bank statement, a prior-year return, a signed letter, a login. Normal window is genuinely several days, because there’s an errand between your message and their reply. Follow up after five business days, and make the ask smaller rather than louder. Which of the six reasons a document is actually late applies here, and what a chase should say in each case, is worked through in managing client document requests without the chasing.
A decision that needs somebody outside the thread. A partner, a spouse, a board, a client of theirs. Normal window is a week or more, and chasing leaves the meeting they’ve yet to have exactly where it is. Follow up after a week, and ask when they expect to know rather than asking again for the answer.
A proposal or a scope document. Shorter than instinct says, because a proposal decays. Follow up after three business days, then again after a week. Following up on a proposal takes the sequence further.
Actions somebody agreed to in a meeting. Send the recap the same day. Follow up two business days after the date they gave, counting from that date rather than from the meeting. Where the recap named a date, you’re covered. Where it named none, fix that before the next meeting rather than chasing it now.
A meeting time. This one is a scheduling problem. Two people trading availability is a loop that became a thread, and scheduling without the email chain is the repair.
When the date belongs to somebody else
Everything above assumes the interval is yours to choose. Often it belongs to a date instead, and then the arithmetic replaces the judgment entirely.
The clearest example is an IRS notice. On a CP2000, its notice of underreported income, the instruction is to “respond within 30 days of the date of the notice or 60 days if you live outside the United States”. The clock runs from the date printed on it rather than the day it landed on anybody’s desk. Miss it and the consequence is specific: “if we don’t hear from you by the response date on the notice, we’ll send you a Statutory Notice of Deficiency” (checked September 6, 2026).
Suppose you need two documents from the client before you can answer. Those thirty days hold your drafting and review time, plus mailing, plus a margin for the document that turns out to be the wrong year. Subtract all of that from the response date. What’s left is the client’s window, and it’s usually about ten days. So the first ask goes out today, the follow-up goes out on day four, and by day seven you’re calling rather than writing.
That pattern generalizes to anything with a fixed end: a filing date, a closing, a funding deadline, a renewal that lapses. Work backward from the date, subtract your own processing time honestly, and the follow-up schedule falls out as subtraction. You’re reading off when it becomes too late, rather than deciding when it becomes acceptable to chase.
One consequence of that is easy to miss. When the deadline sits inside the normal window, the choice is already made for you. Send the request and the date you’ll come back on in the same message, so the second one arrives as a kept appointment rather than as a surprise.
What resets the clock and what only looks like it
Between the first message and the follow-up, things happen that feel like progress. Some of them genuinely change the interval. Some only change how you feel about it.
A date from them resets it, and hands it over. If they say “next week”, the interval is theirs now. Follow up the day after the week they named, and open by quoting them. People answer their own commitments much more readily than they answer your requests.
A holding reply resets it partially. “Got it, I will look this week” moves the clock to the length of what they promised, which leaves it well short of your original interval. It’s also worth more than it looks, because somebody who has replied once has the thread in mind.
An out-of-office adds days on top. Covered above, and worth saying twice because the mistake is so common: come-back day plus one.
A read receipt or an open notification leaves the interval where it is. Knowing a message was opened tells you it was opened. Whether it was read, understood, or workable that morning stays as open a question as it was before. Treat it as a starting gun and you produce chases that feel accusatory for a reason nobody can name.
Your own anxiety moves the clock most often, and has the least right to. If you find yourself recalculating the interval downward on a Tuesday afternoon, the thing that changed is you.
Why their silence feels louder than it is
That last point deserves more than a bullet. The feeling has been measured, and it tracks your own stake in the answer rather than anything the other person did.
Kalman and Rafaeli ran the experiment in Online Pauses and Silence, published in Communication Research in 2011. Fifty-five graduate students in an MBA program at an Israeli university, average age 36, each read one version of a short scenario about a job candidate. The versions varied two things:
- whether the candidate had made a good or a poor impression
- how the candidate answered an email, either the next day, after two weeks, or not at all in over a month
Silence was expensive. Among participants shown the silent candidate, the share who would recommend him fell from 57 percent to 19 percent. A two-week delay pushed the same way, though it fell short of significance on its own.
The result that matters here is the interaction. When the candidate had made a good impression, the two-week reply and the silence both read as violations, and evaluation, credibility and the relational measures all dropped. When the candidate had made a poor impression, the identical delays barely registered. The authors found “no significant difference between the expectedness of the three response latencies”, and concluded that participants “did not perceive a late response or silence from a low reward candidate to be surprising”.
The same silence, over the same number of days, carried a completely different meaning depending on how much the person waiting wanted the answer.
That’s fifty-five business school students in one country, responding to a written scenario rather than to a real inbox, and the authors say plainly that it’s a convenience sample. Treat it as direction rather than as a rate. The direction is enough: the loudness of a silence is a property of the listener.
So the interval that feels right on day four is shaped by how much you need the answer. It pushes you to chase your important threads too hard and to leave your routine ones alone. That’s exactly backward, since the routine ones are the ones that go quiet while nobody’s watching.
The fix is procedural rather than emotional. Pick the interval when you send, while you still know what you asked, who you asked, and what depends on it. That judgment beats the one you’ll make on Tuesday afternoon, and it’s the one moment when the waiting is still ahead of you.
What Point does with the date you named
There’s a mechanical reason people fall back on a flat rule of a week. The date you’d have chosen at the moment of sending has nowhere to live. Your sent folder is ordered by when you wrote rather than by when the answer is due, and what you actually want to find is an absence, which is the one thing a mailbox has never been built to search for. Where the whole list of what you’re owed should sit is tracking what you are owed.
Point is a full email client, so your half of the exchange and the missing half of it live together in one place. When the next move is somebody else’s, you say so in ordinary words as you send. The wait becomes a dated task alongside the rest of your workload, carrying one link back to the conversation it came from. On the day, if the thread is still quiet, it comes back up your feed with a prompt on it.
The part that matters for timing is that the date travels with the context. You set it in the one minute when you know that this client takes a week, that the document needs pulling from a bank, and that the deadline is the 30th. Come the day, the thread arrives with all of that attached, so you read it rather than rebuild it.
It’s conditional rather than scheduled, which is the difference between a follow-up and an alarm. If the reply lands first, the thread is live again and the prompt stays away. Where a wait was settled somewhere else entirely, on a call or in a meeting, Point can flag the task as looking done and show you what made it think so, then wait for you to say yes. And when the thing you’re waiting on is an event rather than a reply, you can leave a standing request to watch for it. It reports to you once and then stops.
That conditionality is what separates it from setting the thread aside until a date you guessed at, which returns regardless of what’s happened since and is covered in snoozed email.
You set how far Point may go by itself, for each kind of action separately. The lowest setting suggests. The middle one prepares the thing and leaves it for you. The highest lets Point act without checking first. Everything begins in the middle. Everything Point did is written down in order, in plain sentences, and that record is what undo works from, though a message already delivered to somebody else’s server is past recalling. The rest of what Point does with an inbox is on the benefits page.
Writing the chase itself stays with you, and that’s the right division. The interval is arithmetic. The sentence that restarts a stalled conversation usually has to name something only you were in the room for.
Common questions
How long should I wait before following up on an email?
Two business days for a question somebody can answer from memory. Five for a request that sends them looking for a document. A week for a decision that needs somebody outside the thread. Adjust against how fast that person normally answers you rather than against a general rule. The useful interval is roughly twice their usual turnaround, which is late enough that the ordinary reply has already had its chance and early enough that the thread is still warm.
Is two days too soon to follow up?
Two days is nearly always fine, and the measured reply curves are the reason. More than nine in ten replies that arrive at all arrive within a day, so by the second business day the ordinary route has been given its full run. What makes a two-day follow-up feel pushy is usually the wording rather than the gap. A message that asks for a date, rather than repeating the ask, lands very differently at the same interval.
Should I count weekends and holidays when I am waiting for a reply?
No. In one large enterprise archive, mail sent on a Sunday took 13 times longer to be answered and mail sent on a Saturday 30 times longer, which is a measurement of how many people are away from their desks. Count business days, start the clock from the next working morning if you sent late on a Friday, and add any days an out-of-office names on top rather than restarting from their return.
How long should I wait to follow up on a proposal?
Shorter than for most client email, because a proposal loses value while it sits. Three business days for the first follow-up, then a week, then two. Ask for the process rather than the decision: who else needs to see it, and when they expect to have looked. That gives you something to refer back to next time, which is the difference between a follow-up and a repetition.
How long before I give up on getting a reply at all?
The research threshold is ten times the person’s normal turnaround, at which point a reply is more than 95 percent unlikely ever to come. For someone who usually answers within a day, that’s around two weeks. Reaching it leaves the thing you needed exactly where it was. What it tells you is that the channel has failed rather than the person. So the next move is a different one: a call, a different person at the same firm, or a message that closes the loop from your side and makes it easy for them to say no.
Does following up quickly make me look desperate?
The speed on its own is rarely the problem. A reader judges the pattern across several messages rather than the gap after one, and the thing that reads badly, at any interval, is a third message that repeats the first with no warning. The cheapest protection costs one sentence on the original: say when you’ll come back if the thread stays quiet. A follow-up somebody was told to expect arrives as you keeping your word rather than as you losing patience.
The intervals and what moves them
Replies are front-loaded far harder than instinct suggests. Most arrive well inside the average, because a few very late ones pull the average up, and by the second business day the ordinary route has had its chance. So the useful interval sits at about twice how long that person normally takes with you. The point where a reply stops being likely at all sits around ten times that.
Then adjust for two things and only two things. What you asked for, since a request that sends somebody on an errand runs genuinely slower than a question they can answer from memory. And whose calendar governs, since a filing date or a notice deadline turns the whole schedule into subtraction and makes the decision for you.
Everything else is a matter of counting honestly. Business days rather than calendar days, from the morning they could plausibly have read it, with their days off added rather than substituted. And pick the number on the day you send, because the version of you waiting on Tuesday afternoon has worse information and stronger feelings than the version who knew what the ask was worth.
What you spend the wait on differs by trade. A signature an accounting practice needs before it can file has a hard outside date and one route to getting it; a scope decision a consultancy is waiting on has an open end instead, and goes cold. Both patterns, and six more, are laid out by who does the waiting.