A churn rate tells you what already happened. A leading indicator is a change in customer behavior that shows up months before the cancellation notice, early enough that the renewal outcome is still in play. The difference between the two is whether the renewals team gets a warning or a receipt.
The support case curve
Support volume is the clearest example of an indicator that fails when read as a straight line. ORM sees the pattern across its customer base: an account with no support cases is at risk of churn, an account with seven or more cases in the last year is at risk, and accounts with three to five cases, usually tier 2 or tier 3 and not severe, are the least likely to leave. Those middle accounts are engaged, getting support, and generally happy.
Read that way, ticket count stops being a support metric and becomes a usage proxy with an opinion attached. It is also the cheapest signal available, because it already exists in the helpdesk with timestamps and owners on every record.
Silence is the hard one
Quiet accounts generate no alerts, no escalations, and no reason for anyone to open them. They look calm, so they sit at the bottom of the customer success queue and arrive as surprise cancellations.
ORM applies the same reasoning to pipeline. Its view on deal slippage is that the earliest signal is the lack of a signal, meaning no activity, no data changing, and no notes. Retention behaves the same way. Absence of evidence gets treated as evidence of health, and that assumption is wrong often enough to be expensive.
The signals worth instrumenting
- License utilization falling against contracted seats, especially a decline concentrated in one team. - Departure of the executive sponsor who signed, tracked from contact records and job change data. - Admin and power user login frequency dropping while total logins stay flat. - An integration disconnecting or an API key going unused. - Support requests narrowing to a single remaining user while everyone else goes dark. - Skipped business reviews and unanswered outreach across two consecutive cycles.
Turning signals into a score
Weight by measured lead time rather than by intuition. Pull the accounts that churned in the last eight quarters, look backward, and record how many days before the loss each signal first fired and how often it fired on accounts that renewed. An indicator with a long lead time and a low false positive rate earns weight. One that fires in the closing weeks of a term earns almost none, because by then the decision has been made.
The common objection is data quality, and it is usually overstated. ORM's position is that every company believes its data is uniquely bad and blames it for weak forecasting, and that the belief is mostly wrong. Garbage in does not have to mean garbage out. What breaks a model is inconsistency, not imperfection. If the same field is filled the same wrong way every time, the pattern is learnable, and the same discipline that protects net revenue retention also holds for forecast accuracy on the new business side.
Frequently Asked Questions
What is the earliest reliable churn signal?
The absence of activity. ORM treats an account filing no support cases as at risk of churn, on the reasoning that nobody is using the product deeply enough to hit a problem. Silence produces no alerts and no escalations, so the account sits at the bottom of every queue right up until it cancels.
Do heavy support volumes predict churn?
Past a point. ORM's read of its customer base is that accounts with seven or more support cases in the last year are at risk, while accounts with three to five cases, usually tier 2 or tier 3 and not severe, are less likely to churn because they are engaged and getting help. The relationship is a curve, not a line.
How far ahead does a good leading indicator fire?
Far enough that the renewal outcome can still change, which means at least one full renewal cycle ahead of the expiry date. A signal that fires in the closing weeks of a term is a report rather than a warning. Score each candidate indicator by median lead time before you decide how much weight it carries.
Our CRM data is a mess. Can we still score churn risk?
Yes. ORM's position is that every company believes its data is uniquely bad and that this is mostly false. Inconsistent data breaks predictions. Consistently imperfect data does not, because a model can learn the pattern in the noise as long as the noise stays stable.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like leading indicators of churn into prescriptive action for your team.
Schedule a Demo