Retention-desk activity moves months before reported churn does, and reseller commentary moves ahead of that. Reported churn is a lagging indicator by construction, because it counts subscribers who have already left — which is why a quarter can look stable in the print while the leading signals have already turned.
Why does churn reach reporting last?
Because it counts subscribers who have already gone, and most contracts have notice periods. By the time churn appears in a quarterly number, the decisions behind it were made one or two quarters earlier.
That is a definitional property rather than a disclosure failure. Churn is a completed event and the interesting question is about events in progress — which is also what a retention figure hides.
That construction also explains why churn guidance is unusually reliable and unusually late. A company forecasting churn is forecasting something largely already determined, so the number tends to land close to guidance — which reassures the market about a quarter whose underlying behavior changed some time ago. Accuracy against guidance and usefulness as a signal are different properties.
Which signals lead reported churn?
Retention-desk volume and the aggressiveness of save offers, followed by channel and reseller commentary. Both are observable to people who worked in or alongside the business.
The save-offer signal is the sharper of the two, because it responds immediately: more customers asking to leave produces more offers before it produces more departures.
Both signals need a baseline to be readable. A save-offer level means nothing in isolation and a great deal against what the same desk was authorized to offer two quarters ago, which is why these questions work best asked of someone who has watched the same operation over time rather than of a single recent leaver.
What does the retention desk see?
How hard the company is working to keep customers and what it is willing to pay. A rising average save offer is a churn signal even while the churn number is flat, because it means the underlying intent has shifted.
Desk staff also know which segments are calling, which is information the aggregate number destroys.
The desk also sees the reason mix, which the reported number collapses entirely. Customers leaving on price, on service and on a competitor's promotion behave differently afterward: price-driven churn responds to a counter-offer, competitive churn frequently does not, and service-driven churn returns only if something structural changed. A rising number with a shifting mix is a different problem from a rising number with a stable one.
What do resellers see first?
Connection and disconnection volumes across their own base, and competitive offers arriving in the market. Where a business sells through channel, the channel sees the switching before the operator does.
That makes resellers the earliest signal available and the noisiest, since their own commercial position colors what they report.
Reseller commentary also arrives with a bias worth correcting for. A reseller losing share will describe the market as difficult and one gaining will describe it as rational, and both are describing their own position as much as the operator's. Reading two or three against each other is what converts the noise into a direction.
How should a model use this?
To adjust timing rather than level. Leading signals reliably indicate that a churn change is coming and roughly when; they are unreliable for magnitude, and models that read them as a rate overreact.
Used as a timing input they are genuinely valuable, because the market usually prices the reported number and not the one in formation — which is the window a pre-earnings check works in.
Used carefully, the signals also tell you what will not happen. A quarter in which retention activity has been flat is reasonable evidence that the reported churn number will not surprise, and knowing a number is unlikely to move is worth something to anyone positioned against it. Disconfirmation is the more reliable output of leading-indicator work.
Timing adjustments are also easier to defend than level ones. Saying that a change is arriving a quarter earlier than the market expects is a claim about sequence that the next print will settle, while a claim about magnitude usually cannot be resolved by any single report.