You can usually feel that your DM sales are leaking somewhere. Asking where tends to produce a shrug, or a vague answer about needing to follow up more, because the evidence is spread across months of threads and nobody reads back through those for fun.
Search the question and you land on software. The answer offered is a platform that watches your conversations and draws you a chart of where qualified leads fall out, which is a real product and a sensible thing to buy at a certain size. It is a strange first move for someone with a few hundred conversations, though, because the chart shows up before you have reread a single one of them.
Doing it by hand once teaches you more than a year of charts would, and not because manual work builds character. The stage where people stop is attached to a specific sentence you wrote, and the only way to find that sentence is to look at the moment they stopped.
The five places a DM deal stalls
Every conversation that did not end in money ended at one of five points, and they are not interchangeable.
| Stage | What happened | What you know about them |
|---|---|---|
| 1 | You wrote, nothing came back | Nothing, and possibly not even that they saw it |
| 2 | They replied, then went quiet before any number | Their words and their situation |
| 3 | They got the price, then went quiet | All of the above, plus their reaction to the number |
| 4 | They agreed to a call that never got on a calendar | They wanted it enough to say yes once |
| 5 | The call was booked and they did not show | Everything, except why they stayed home |
The reason to split it this way is that each boundary belongs to a different thing you wrote. Stage one is your opener and where the conversation came from. Stage two is whether your questions were answerable. Stage three is how you delivered the price, not the price itself. Stage four is the move you make between a yes and a calendar entry. Stage five is everything that happens in the gap between booking and the call.
You will notice there is no stage for "they said no." A no is an outcome, not a leak. It costs you nothing except the time already spent, and a month with more clean nos in it is usually a better month than one full of silences.
Why there is no benchmark worth copying
The first thing people want after counting is something to compare against, and this is where most articles hand over a percentage. We are not going to, and that is not false modesty.
Stage rates depend on things that differ completely between two coaches who look identical from the outside. Where the conversations came from matters enormously: threads that start from a referral behave nothing like threads that start from a Reel that traveled past your usual audience. So does your price, your niche, how long your offer takes to explain, and whether you opened the conversation or they did. Two people with the same stage three rate can be running two businesses with nothing in common.
The only number that means anything is your own, measured against itself. Your first tally grades nothing, it simply gives you something to hold the next one up against. It becomes useful the second time you run it, when you can see which boundary moved after you changed something, and it becomes useful in a different way immediately, because one of the five falls is almost always visibly worse than the rest.
Credit where it is due to the platforms that sell the dashboard version of this: continuous measurement beats a monthly count, it never forgets, and at a few thousand conversations a month doing it by hand stops being realistic. The thing a dashboard cannot give you the first time is the reading. You get a number for stage three without having reread the five conversations that died there, which is where the actual answer was sitting.
The tally: one month, one sheet, under an hour
Pick a month that is finished and has had time to breathe. If it is late November, count October. A conversation that started three days ago has not dropped off anything yet, it is simply young, and counting it as a loss will make your numbers worse than your business.
Then build the list. One row per person, five columns, and in each row you mark the furthest stage that conversation ever reached, as of today rather than as of the last day of the month.
The rule that trips people up is what to do with a conversation that spans two months. Someone wrote to you on October twenty-eighth, you talked, and the call happened in the second week of November. That conversation belongs to October, because the month a conversation starts is the only date that never changes, and it is scored at stage five, because you score where it got to and not where it was when the calendar flipped. Score by activity month instead and you will double count half your inbox.
Second rule: score what is written down rather than what you remember feeling. It is tempting to mark a warm exchange as a yes to a call because you remember it going well. Look at what is actually written. Something like this is not a yes:
Yeah that sounds good, send me the details when you get a chance
That is stage two until a specific time has been discussed, however promising it felt at the time.
Under an hour is realistic because you are not reading entire threads. You open the ones where you wrote something, scroll to the end, and ask which of the five boundaries it died at. With a couple of hundred names that is a few seconds each. A spreadsheet is easier than paper only because it adds the columns for you.
When the columns are added, look at the fall between each pair of consecutive stages, in plain counts before you turn anything into a percentage. The biggest fall is the sentence to go rewrite. If you want the count of follow-ups each of those piles deserves afterward, that is a separate decision and we worked through it in how many follow-ups before giving up.
Two of the five stages are not what they look like
Stage one and stage two both lie to you, in opposite directions, and knowing how keeps you from acting on noise.
Your "never replied" pile is contaminated, because some of those people never had your message in front of them. On Instagram a message from someone a person does not follow arrives as a request rather than in their inbox, and Meta's own Instagram Platform documentation adds a detail most people miss: all message requests answered using a third-party app get moved to the General folder, and folder information is not available to connected apps at all. So neither you nor any tool you plug in can tell, from the outside, whether a thread was ever somewhere the person looks. Treat stage one as a floor. And if most of that pile is people who do not follow you, you are not measuring follow-up at all, you are measuring something closer to cold outreach, which has its own economics.
Stage two lies the other way, because "left on seen" feels like a measurable event and is not one. Instagram lets people turn read receipts off, for every chat or for specific ones, as Meta announced in its own DM update. LinkedIn has the same switch, and its help page on read receipts and typing indicators states plainly that if a recipient has turned the setting off, you will not see when they have read your message. A thread with no checkmark is not a thread that went unread, so build stage two out of the last message that exists rather than out of a receipt. What to do with the ones you genuinely know were read is a different question, covered in when a lead leaves you on seen.
The reading mistake that makes your numbers lie
There is one way to get this exercise completely backwards, and it is the first thing almost everyone does with their second tally: comparing a stage count to last month's stage count when the number of new conversations changed underneath it.
Imagine October brought you 40 new conversations and 12 of them reached the price. November, on the back of one post that traveled further than usual, brought 110 new conversations and 18 reached the price. The stage three count went up by half. Read it that way and you conclude the pricing conversation is improving, when the share of people getting that far fell by nearly half. The numerator moved because the denominator did, and the business got worse in the month that looked better.
The fix is to compare falls rather than counts: for each boundary, what share of the people who reached it went on to the next one. Those shares are stable against volume, which is exactly what makes them comparable month to month. Counts are still worth writing down, because they tell you how much money is sitting at each boundary, but they are not the thing you compare.
Even done correctly, be careful how hard you lean on a single month. Pew Research Center, an organization whose entire job is reading numbers off samples, will not report findings for a group with an effective sample size under 100, because below that the margin of error passes 10 percentage points, and it declines to publish estimates at all below an effective sample size of 50. Most coaching businesses start fewer conversations in a month than that. So treat a single month as a hint: act when a boundary moves the same direction for three months running, or when one of the five falls is so much worse than the others that no amount of sampling noise explains it.
What each stage tells you to rewrite
Once you know which boundary is eating the month, the work narrows considerably.
A heavy fall at stage one sends you to your opener and to your sources, in that order. The opener is the cheaper thing to change, but if the pile is mostly strangers, no opener rescues it.
A heavy fall at stage two usually means your messages were not answerable. The test is whether a given message could have been sent to anyone in your inbox. Compare these:
Hey! Just checking in, let me know if you had any other questions.
You said the mornings are the part that falls apart. Is that still the thing you'd fix first?
The second one requires having read the conversation, and it can be answered while someone waits for coffee, which is the only standard that matters.
A heavy fall at stage three is about delivery, not about being too expensive. Most of the damage is done in the message that carries the number, and how to answer the price question in DMs goes through that in detail. When the response to the price is a stated delay rather than silence, you are in different territory and the move is covered in what to do with a lead who stalls.
A heavy fall at stage four is almost always mechanical rather than emotional. People agreed, then were handed a booking page and a task, and the task never got done: why sending your booking link kills your calls is the single most common cause of a pile at this boundary.
A heavy fall at stage five points at the gap between the yes and the call. A booking made a day before holds far better than one made two weeks out, and a confirmation that says what will happen on the call and offers an easy way to move it tends to convert the silent no-show into a reschedule.
Change one boundary at a time. If you rewrite your opener, your price message and your booking move in the same month, your next tally will tell you something changed and nothing about which of the three did it.
What Ficiel does with this, and what it does not
The tally is yours. Ficiel does not hand you an industry benchmark for any of these five stages, because we do not have one worth handing over, and a made-up number is the thing this whole exercise is meant to replace.
What Ficiel removes is the archaeology. It brings Instagram, WhatsApp and LinkedIn into one list, rereads your history (24 months of it on the Essentiel plan), and surfaces on its own the conversations where you spoke last and got nothing back, with their context attached, instead of asking you to scroll through weeks of threads to find them. Each contact carries a stage, so the second time you run this count you are reading a list rather than reconstructing one.
Kord is Ficiel's AI assistant: it rereads your conversations, finds the leads who went quiet, drafts follow-ups in your own voice, and a human always approves before anything is sent. For this subject the approval is the point, because the message that fixes a stage three leak is one you decided to write after rereading five conversations, and no assistant has that reason until you give it.
What it does not do is fire a sequence because a timer ran out, or decide on its own that a conversation is dead. Ficiel's Essentiel plan costs $117 a month, with money back within 10 days.
FAQ
What is Ficiel?
Ficiel is an AI inbox for coaches and setters: it surfaces old Instagram, WhatsApp and LinkedIn conversations worth following up, drafts the message, and nothing is sent without your approval. For this question it mostly removes the scrolling: the threads where you spoke last and got nothing back are already gathered, with their context, so counting where people stopped takes an hour instead of an afternoon.
How long does the tally really take?
Under an hour for a few hundred conversations, once you accept that you are not rereading them. You open each thread where you wrote something, look at the end, and mark which of the five boundaries it stopped at. The part that takes real time is afterward, when you reread the handful of conversations that died at your worst boundary.
What counts as a conversation for this?
Any thread where you were trying to sell something, including ones the other person opened. Keep out conversations with existing clients, friends and anyone who messaged you about something unrelated, because they are not leads and they will flatter your stage one number. Count each person once, in the month their sales conversation started.
What is a good reply rate in DMs?
Nobody can tell you, and anyone who gives you a figure is describing a different business. Reply rates swing with where the conversations came from, what you sell, what it costs and who opened the thread. Your own previous month, measured the same way on a comparable number of new conversations, is the only comparison that carries information.
Which stage should I fix first?
The one with the biggest fall, measured as a share rather than a count, unless a smaller fall sits much later in the sequence. People who reached stage four cost you far more to replace than people who stopped at stage one, so a modest leak just before the call is often worth attacking ahead of a larger one at the top.
Can I do this with inbox labels instead of a spreadsheet?
Yes, and it is better if you will actually keep it up, since a label applied as the conversation happens is more accurate than a stage assigned from memory weeks later. The catch is that labels tell you where things stand today, not where a given month's conversations ended up, so you still need to note the month each thread started if you want to compare two of them.
