the cheapest customer to save.

leaving is a crossing, not an event

The timestamp in your database says a customer churned on a Tuesday at 14:32. That is when the cancel button was clicked. The decision was not made then. It was made over weeks, sometimes months, as the customer moved from settled to unsettled to gone. We argued in your churn model predicts the past that the crossing is the least informative part of the event. This note is about the part before it: the transition state, and why it is the single best place to spend a retention dollar.

Picture the landscape again. Staying is a basin. Leaving is a basin. Between them stands a barrier, and at the top of the barrier lies a narrow region where the customer belongs to neither side. Chemists call it the transition state, and they have known for a century that it is where reactions are decided. Not in the basins. At the top.

the physics of the window

The transition state has one property that matters commercially: it is the point of maximal sensitivity. Deep in a basin, restoring forces pull the customer back toward equilibrium. A nudge changes nothing, because the basin absorbs it. At the top of the barrier the forces are balanced, and balance means a small push decides which way things fall. This cuts both ways. A minor snag, a competitor ad, or one clumsy support reply can finish the crossing. And the same sensitivity means a small, well aimed intervention can send the customer back down the slope they came from.

Then comes the asymmetry. Once the customer has crossed and settled, they form a new equilibrium. New subscription, new routine, setup costs sunk on the other side. Bringing them back now means paying the full barrier height again, against a competitor who is now the incumbent. This is why win-back campaigns convert poorly and expensively.

A win-back campaign is not arguing with a decision. It is fighting a new equilibrium.

what a customer in transition looks like

If the window is that valuable, why does almost nobody spend there? Because customers in transition are hard to see with the instruments retention teams run. Usage often holds steady through most of the crossing. Habit keeps the metrics green while commitment is already moving. These are the low score, high fragility customers from the upper left quadrant of the previous note, invisible to a behavioral model by construction.

But transition has a signature, and it shows up in language before it shows up in behavior. The destination enters the sentence: complaints stop being complaints about you and become comparisons to someone specific. Churn has a grammar, source, destination, trigger, and the crossing begins when the destination gets named. Friction that was silently absorbed for years suddenly gets articulated. Questions change shape, from how do I do this to is this still worth it. Each signal is ambiguous on its own. What is not ambiguous is density. When the share of a segment speaking in transition grammar starts climbing, that segment is moving up the slope. This is why we count density, not presence. One customer naming a competitor is an anecdote. Four percent of a cohort naming the same competitor with the same trigger is a migration in progress.

where retention budgets actually spend

Now look at where retention money goes. Two places, mostly. Blanket discounts and loyalty pricing aimed at the stable base, which pays customers deep in the basin to do what they were doing anyway. The basin absorbs the discount the way it absorbs everything else, and the cost comes straight out of margin. And win-back campaigns aimed at the departed, which is the most expensive force per customer anywhere in the landscape. The window in the middle, where force is cheapest, gets whatever is left. Usually nothing, usually because nothing in the stack can see it.

The intervention itself is often not even money. A customer at the top of the barrier is usually there because of a named friction, and the grammar tells you the trigger. Fix the snag they named. Acknowledge it specifically. Time an outreach to it. A small force, applied where the forces are balanced, does the work that a thirty percent win-back discount fails to do three months later.

measuring the window at population level

Limits first, as usual. Public discourse will not tell you that a named individual is mid crossing this week. That resolution belongs to your first party data, if you instrument for it. What the population level reading gives you is earlier and structurally different: which segments are accumulating transition state density, which trigger is doing the pushing, and how wide the window is. Window width varies by category, the same way complaints have a half-life: transitions have a dwell time. A budgeting app on an annual plan dwells near the top for months, a wide window. A coffee subscription crosses between two deliveries. Knowing your category's dwell time tells you how fast your intervention loop has to run for early churn detection to be worth anything.

The instrument is the same one that runs through this series. Cluster the category's discourse, stratify it by regime, and watch the density of transition grammar per segment over time. Read that curve as an early warning instead of an autopsy, and the alert fires while the customer is still at the top of the barrier, where saving them is cheap, instead of after the cascade, when the dashboard finally turns red. The window is not hypothetical, and it is measurable at scale: dfmchn's Netflix price increase stress test flagged a roughly 14% at-risk cohort from public signal alone and found most of it still reversible. That reversible share is a population standing at the top of the barrier, not past it.

---

Churn prevention is not about predicting who leaves. Prediction arrives with the autopsy. It is about knowing where the window is and arriving while it is open. The cheapest customer to save is the one who has not finished deciding, and the instrument that finds them reads language, not logs.