The decision, stated properly
Churn shows up as a surprise a couple of times, and the response is reasonable: nobody owns the client after the sale, so make it someone's job. Hire for customer success.
The question underneath is what that person will actually work from. Retention is not primarily an effort problem — most founders and most teams genuinely care and would happily attend to an account they knew was at risk. It is a detection problem. And a hire adds capacity for attending, not capacity for detecting.
So this is a decision about sequencing rather than about whether to ever hire. A CSM with a health signal is a strong hire. A CSM without one is a very expensive way to be attentive to whoever shouts.
What a customer success hire does well
The single biggest thing they fix is ownership. In most founder-led businesses no one is responsible for the client between projects, and unowned things do not happen. Naming a person changes that immediately, and it is worth a lot.
They also do work that genuinely cannot be systematized. The check-in call that surfaces a concern nobody wrote in a ticket. Noticing that the client's new stakeholder has different priorities. Reading the tone of a message and deciding it needs a phone call. Those are human judgments and no instrument produces them.
And they create a relationship independent of the founder, which is one of the few durable protections against founder dependency in the client relationship. A client who has a real relationship with someone other than you is a client who does not leave when your attention moves.
For a business with a working health signal and accounts that need active management, this hire is correct and often overdue.
Where it structurally breaks
The break is in what they can see. Give a capable CSM twenty-five accounts and no data about which are drifting, and they will do the only thing available: work the accounts that generate contact. That is attention allocated by urgency, which is precisely the mechanism that caused the churn.
Then the reporting problem. Without a health signal, a CSM's status reports are impressions — this one feels good, that one has gone a bit quiet. Impressions are not nothing, but they cannot be trended, cannot be handed over, and cannot be checked. You will be no better informed than before, but you will feel better informed, which is worse.
There is a knowledge concentration risk too. Everything they learn about each account lives with them unless something records it. When they leave, the business loses not just the relationship but the entire accumulated model of every client, and the next person starts from a list of names.
And the failure is slow to detect. Churn is a lagging outcome with a long delay, so a CSM who is not preventing it will look fine for two or three quarters. By the time the data says otherwise, the hire has been in place for a year.
The compounding version is worth naming. A CSM without instruments spends their first year building a private mental model of every account, which is genuinely valuable and entirely unrecorded. The better they are, the more the business depends on their memory — so a successful hire in this configuration quietly recreates founder dependency with a different name on it.
What we build instead
We build the instrument that makes the quiet visible. Concretely: a health signal per account assembled from things the business already produces — days since a human made contact, reply latency compared to that client's own baseline, work items past their promised date, scope of conversation over the last quarter. None of that is exotic and all of it is currently uncaptured.
Then a review rhythm that catches drift before it becomes a decision. Every account looked at on a defined cadence, with a defined action for each warning sign, so a signal produces a person doing something rather than a founder feeling uneasy.
Then the renewal architecture, because renewals are the single most engineerable moment in a client relationship and almost nobody engineers them: the date surfaced in the operating rhythm rather than living in a contract, a review conversation ninety days out while it is still a review rather than a negotiation, and a record of the year's delivery that does not need a week of reconstruction to assemble.
And the touchpoints that should not depend on anyone's memory — the check-in that fires, the milestone that gets acknowledged, the thing they mentioned in March that comes back up in June. Automate the reminder, but write the message like a friend. The system's job is to make sure the moment arrives; the message is the part a person must still write, and protecting that is the point.
The output is that attention stops being allocated by whoever is loudest. Anticipation is the highest form of care — and anticipation is not intuition, it is a pattern you already know, written down and pointed at the accounts that make no noise.
How to tell which one you need
Ask one question and answer it honestly: can you name, right now, which three of your accounts are most at risk, and say what evidence you are using? Not a feeling — evidence.
If you can, you have a signal, formal or informal, and your constraint is capacity to act on it. Hire the customer success manager. They will arrive with something to work from, their reports will mean something, and they will be measurable. A business in that position that builds more instrumentation instead is adding resolution to a picture it can already read.
If you cannot, a hire will inherit the same blindness and the same twenty-five accounts. Build the signal first — it is weeks rather than quarters, it costs far less than a salary, and it frequently reveals that two accounts needed attention rather than twenty-five, which changes the hiring question entirely.
There is a third case worth naming: enough accounts that even with a perfect signal nobody has the hours. Then you need both, and the order still holds, because the signal is what makes the hire's first ninety days useful instead of exploratory.
Retention fails on detection, not effort, and a hire adds capacity to attend rather than capacity to detect. Name your three most at-risk accounts and the evidence — if you cannot, build the signal first.