The customer does not decide to leave on the day they cancel.
They decide weeks earlier. Sometimes months. The cancellation is administrative confirmation of a decision that was already made — quietly, privately, without any signal that appeared in your dashboard, triggered an alert in your CRM or generated a flag in your governance meeting.
By the time the churn event registers, the conversation that might have changed the outcome is no longer available to you. The renewal call, the retention offer, the service recovery — they arrive after the decision, not before it.
I call this the Pre-Exit Window: the period between the moment a customer decides to leave and the moment that decision becomes visible to the organisation. In most organisations, this window is 60 to 90 days long. And almost nothing that happens during it reaches the people who have the authority to act on it.
85% of customers who left a provider say they would have stayed if their problem had been addressed in time. (Netigate, 2025)
Closing that window — or at least narrowing it — is not a technology problem. It is a governance problem. And it starts with understanding what the signals look like before they disappear.
Source: Netigate 2025 customer churn analysis. Behavioral research consistently shows churn signals appear 60-90 days before cancellation (industry benchmark across SaaS and service sectors).
Why organisations miss the signals
The signals exist. The governance structure to act on them does not.
The signals that precede customer exit are not invisible. They exist in operational data, in contact patterns, in digital behaviour, in frontline conversations and in the gap between what the customer was promised and what they are receiving.
What makes them difficult to detect is not their absence. It is where they appear — and what the organisation is set up to look for.
Most churn analysis is retrospective. It asks: which customers left, and what did they have in common? This is valuable for building predictive models. But it does not help the customer who is currently in the Pre-Exit Window — whose signals are present now, in the operational data of the organisation, but have not been connected to a governance process that would enable intervention.
The reason for this failure is structural. The signals of impending exit are distributed across multiple functions — service contacts, digital interactions, billing data, renewal timelines — and no single function has both the visibility and the authority to connect them and act on what they reveal.
The result is a pattern that most organisations recognise in retrospect: the customer who seemed fine until they weren't. Whose NPS score was positive three months ago. Whose last service interaction was rated satisfactorily. Who gave no indication, in any individual interaction, that they were managing an exit.
They were not hiding. The signals were there. The organisation simply was not looking for them in the right places — or connecting what it found to someone who could act.
The six signals of impending exit
None sufficient alone. Together, they form a pattern that is consistent across industries.
Signal 1: Declining engagement without complaint
A customer who was once active becomes less active. Login frequency drops. Usage of core features decreases. Interactions that were regular become occasional. The customer is still technically a customer — they have not cancelled, they have not complained — but their relationship with the product or service is quietly contracting. Research published in the Journal of Service Research (2024), analysing over 840,000 customer interactions, found that communication cessation — silence — is a stronger predictor of customer defection than complaint frequency, across both B2B and B2C contexts. Organisations that track engagement data at the individual customer level can detect this signal early. Most track aggregate engagement across the customer base and miss the individual trajectory that precedes departure.
Signal 2: Contact pattern changes
A customer who contacts the service team more frequently than usual, or whose contact reason shifts — from transactional queries to complaints, from self-service to agent escalation, from a single issue to multiple unrelated issues — is exhibiting a contact pattern that consistently precedes exit. The frequency shift signals frustration accumulating. The reason shift signals a deteriorating relationship with the product or service. The escalation pattern signals a customer who no longer trusts routine channels to resolve their problems. Most contact centre systems track volume and resolution time. Few track individual customer contact trajectories over time in a way that would surface this pattern before it becomes visible in churn data.
Signal 3: Reduced spend, scope or usage
In B2B contexts, a customer who reduces their order volume, declines a renewal of an additional service, removes users from a platform, or quietly reduces the scope of their engagement is exhibiting a commercial signal that almost always precedes full exit. In B2C and subscription contexts, the equivalent signals are plan downgrades, feature removal, reduced order frequency or cancellation of add-ons. In both cases, the customer is not leaving yet — they are reducing their exposure, testing the experience at a lower level of commitment before making the final decision. The reduction in scope is a reversible decision. The cancellation that follows, if the signals are not detected and acted on, is not.
Signal 4: Reduced responsiveness
A customer who stops responding to emails, declines calls, delays renewal conversations or becomes difficult to reach is not simply busy. They are managing the relationship at arm's length — maintaining just enough contact to avoid a direct confrontation while they prepare to leave. This signal is particularly prevalent in B2B relationships where the commercial relationship has a defined renewal point. The customer who was responsive six months ago and is now unreachable is almost always in the Pre-Exit Window. The organisation that does not recognise this pattern will arrive at the renewal conversation without any of the context it needs to have a productive discussion.
Signal 5: Frontline signals without escalation
Frontline employees — service agents, account managers, support specialists — often know a customer is at risk before any data system detects it. A tone change in conversations. A comment that reveals frustration with a competitor offering. A question that only makes sense if the customer is evaluating alternatives. These signals exist in the frontline relationship. They almost never reach a governance process in time to enable intervention — because there is no structured mechanism for frontline employees to flag at-risk customers in a way that triggers a response from someone with the authority to act. This is the Frontline Intelligence Gap applied to churn prevention. The signal exists. The organisation cannot access it.
Signal 6: Experience failures without relationship recovery
A customer who experienced a significant service failure — a billing error, an onboarding breakdown, a promise that was not kept — and whose issue was resolved at the transactional level but never acknowledged at the relationship level, is carrying unresolved dissatisfaction that most measurement systems will never detect. The transactional resolution closes the ticket. It does not close the relationship wound. The customer gave the interaction a satisfactory rating because the agent was helpful. But the failure that generated the contact — the process gap, the broken promise, the repeated friction — was never addressed at the structural level. And the customer knows it. This signal is the most insidious because it appears nowhere in the data. The interaction was resolved. The survey was positive. The account looks healthy. The customer is in the Pre-Exit Window.
Why detecting signals is not enough
Signal detection without governance produces awareness without action
Identifying the signals of impending exit is necessary. It is not sufficient.
The signals must connect to a governance process — a named owner, a defined intervention pathway, a timeframe for action. Without that connection, signal detection produces awareness without action. The at-risk customer is identified. Nothing changes. The Pre-Exit Window closes. The churn event registers.
Most organisations that invest in churn prediction face exactly this failure. The model identifies at-risk customers. The output goes to a report. The report is reviewed in a monthly meeting. The meeting generates a follow-up action. The follow-up action is assigned to the account team. The account team attempts contact. The customer has already left.
The timeline of the process is longer than the Pre-Exit Window it is supposed to close.
Effective churn prevention requires three structural conditions — each with a concrete first step.
1 Named ownership at the account level
A specific person — not a team, not a function — who is responsible for the relationship with each customer and who has the authority to act when a risk signal is detected. In B2B contexts, this is often the account manager. The structural question is whether that person has both the visibility to detect the signals and the authority to intervene without committee approval.
2 A signal aggregation mechanism
A process that connects signals from multiple sources — engagement data, contact patterns, commercial signals, frontline observations — into a single at-risk view of the customer. No individual signal is sufficient. The pattern across signals is what enables reliable early detection. Bain & Company's retention research is direct: a 5% improvement in customer retention can lift profit by 25 to 95%. The lever is there in most organisations. The signal aggregation mechanism is what makes it accessible.
3 An intervention protocol with a defined timeline
A defined response when the risk threshold is crossed: who contacts the customer, in what timeframe, with what authority to offer resolution, and what escalation path exists if the first contact is unsuccessful. The protocol must be faster than the Pre-Exit Window it is designed to close. A process that takes four weeks to reach the customer is not a retention intervention. It is a post-mortem.
The question before your next retention review
Most organisations review churn after it has occurred. They analyse the customers who left, identify the common factors and build programmes to prevent the next cohort from following the same path.
This is valuable. But it addresses the wrong problem.
The customers who will generate next quarter's churn event are in the Pre-Exit Window now. Their signals are present in your operational data today. The question is not what you will learn from their departure. The question is whether you have a governance structure capable of detecting and acting on what is already visible.
For the customers who churned last quarter — at what point were the first signals of impending exit visible in our operational data, and did our governance process detect them before the Pre-Exit Window closed?
If the honest answer is that the signals were present but undetected — you do not have a churn problem. You have a signal detection and governance problem. And the customers currently in your Pre-Exit Window are waiting to find out whether you will solve it before they leave.
If this is relevant to your organisation — share it with the person who owns your customer retention process and the person who reviews your churn data. If they are different people with no shared accountability for the signals between them, that is where to start.
-> If you want to assess how well your organisation detects at-risk customers before they leave, the Customer Churn Risk Test at experiencefirst.lt identifies the six dimensions of churn risk visibility — and shows where your early warning system is strongest and where it has gaps.