A client who is leaving decides weeks before saying so, and in the meantime their behaviour changes: approvals come later, the report stops being commented on, meetings get cancelled with no alternative date offered and somebody new starts doing the talking. All four can be measured against that client's own history, and the answer is decided by looking at margin and capacity, not by producing more.
The decision was already made when the email arrives
Almost no cancellation is a surprise to the client. By the time the message saying they have decided not to renew turns up, that sentence has been written in somebody's head for six to ten weeks, and in that time there have been meetings, approvals and a couple of reports that the agency read as ordinary months.
What makes them read as ordinary is that churn signals are not negative. Nobody complains. An angry client is a client still investing energy in the relationship, and those very rarely leave without warning. The one who is leaving becomes polite, brief and punctual about the bare minimum, which is exactly the behaviour profile no agency investigates, because it does not cause any trouble.
So the useful signals are about rhythm, not tone. You do not read them in what the client says, you read them in how long they take, how often they reply and who they reply through. And all of them share the same awkward property: they mean something only against that same client's earlier behaviour, and against nobody's average.
The four signals, and where they show
| Signal | Where it shows | When to raise the alarm |
|---|---|---|
| Approvals slow down | The approvals board and the pieces parked for rescheduling | Two weeks running above that client own average |
| The report stops drawing comment | No question and no reply after you send it | Two months running with no reaction at all |
| A new contact appears | Who signs the emails and who joins the meeting | The day they appear, not once they settle in |
| Meetings fall through | A cancellation that brings no alternative date | The first time, if it had never happened before |
| Extras stop being asked for | The quotes that are no longer requested | A full quarter inside the retainer and nothing else |
None of the five counts on its own. A slow month of approvals is August; two months without comment on the report can be an internal change of priorities; a new contact may simply be a new hire. What is almost never a coincidence is two of them lighting up at once, and less so when they light up in the order they are written in, which is the order they usually arrive in.
The fifth deserves its own note, because it gets mistaken for good news. A client who stops asking for things outside the retainer looks like an easy client, and sometimes is one. But asking for an extra is an act of trust and of projection: you give additional work to somebody you expect to still be working with next year. When quotes get opened and never answered, and the share link records that they were looked at, the reading is not that everybody is busy.
The fourth has a measurement wrinkle worth settling before you use it. A cancelled meeting and a postponed meeting feel almost identical in memory and share nothing in fact, and telling them apart requires both to be written down somewhere with a date. Once meetings arrive from your calendar or your booking link and stay on the company's record, "we seem to see less of them lately" stops being an impression and becomes a column, which is half of what a CRM is for.
Slow approvals are the first to show up
Of the five, this is the only one that measures itself with no subjective judgement, so it is the one to start with. Every piece waiting for approval has a date it was sent and a date it came back, and the distance between the two moves before anything else in the relationship does.
The board separates what is pending, what is approved, what was rejected and what missed its slot and needs rescheduling. That last column is the one to watch, because it does not fill up with bad pieces: it fills up with pieces nobody looked at in time. A piece whose hour passes while it waits is not published late, it is parked, and a client who accumulates three parked pieces in a month is saying something even while still answering emails politely.
There is also a distinction worth drawing between two things that look alike. A client who approves late and then asks for changes is engaged and slow. A client who approves late and approves everything untouched has stopped looking, and it is the second one that should worry you. Approving everything is a form of disengagement, not a form of trust, and the same board shows it if you watch whether anything ever comes back rejected.
The new contact, and why it announces itself
The most reliable of the five is not a number, it is a name. The day somebody who was not there appears on the thread, or a new person joins the monthly meeting just to listen, the relationship has changed hands even though nobody said so.
That happens for two reasons and both matter. Your contact may be leaving the company, in which case you lose the person who hired you and defended you in the meetings you were not in. Or somebody above them may have decided to review the spend, and the new person is there to assess it. In both cases the correct reaction is the same and it is immediate: introduce yourself, ask what they expect, and treat it as an onboarding rather than as a visit.
Which is why it is worth having every contact, every meeting and every note live on that company's record instead of in the head of whoever runs the account. When the contact changes, the only thing holding the relationship together is that the history is written somewhere another person on your team can reach. How contacts, meetings and opportunities get recorded is covered in the CRM for agencies.
Everything in this article happens in one place in GoFeed.
Try it freeMeasure yourself against your own past, and nobody else's
Churn percentages circulate in this category as though they were a yardstick. They are not. A churn figure depends on who you sell to, how long your contracts run, whether you work with three year old businesses or thirty year old ones and how many clients you have, and comparing yours with another agency's produces calm or panic without producing information. You will not find that number here.
What does work is your own history, and you have to start keeping it before you need it. Four measures are enough and all of them come from data you already have:
- Average days to approval, per client and per month. It is the series that moves first and the only one that leaves no room for interpretation.
- Months of relationship at the end, for every client who has left. If most of them go in month seven, your problem is in month five and not at renewal.
- Renewals over the last four quarters, as a count and not a percentage. On a small book, a percentage exaggerates every case and hides the trend.
- Real margin per client, with the hours as they really were. The client who bills most is almost never the most profitable, and sometimes the one who wants to leave is the one you should let go.
That fourth point is what changes the tone of the conversation. With the four series in front of you, the question of the quarter stops being how to stop somebody leaving and becomes who you want to keep, which is a question that has an answer. The monthly report is where those series get told without sounding like a defence, and how to write one that gets read is in the monthly report as a renewal tool.
The conversation, and when the one leaving is you
When two signals light up, the move is not to produce more or better work. It is an explicit conversation, and it belongs in a meeting called for that purpose rather than in the last four minutes of the monthly one, when everybody wants to hang up. Two questions are enough: what did you expect when we started, and what have you received. The second answer stings more than expected and it is the only one that helps.
One of three things almost always surfaces. An objective that was never written down with numbers, in which case the problem belongs to the start and gets fixed by going back to it, which is what a good onboarding checklist is for. An internal change at the client that nobody told you about. Or a mismatch in expectations that has existed since month one and nobody dared name.
And there is a case where the right answer is to let them go yourself. When the real margin has been negative for two quarters and the scope cannot be cut, keeping that client is not free: it costs exactly the clients you cannot take on while you are busy with this one. That cost can be worked out before you decide, and how many accounts each person on your team genuinely sustains is estimated by the clients per person calculator. Ending a relationship that no longer works, and ending it well, is also the cheapest referral there is.
